A reading of what is actually happening in the market, for those trying to make sense of AI adoption at the organisational level.

Three numbers worth keeping in mind

Base44, a startup launched in January 2025 by an Israeli founder (ex-Unit 8200, Forbes 30 Under 30), reached $1M ARR in three weeks after launch. Six months later, Wix acquired it for $80M in cash. Number of employees at the time of acquisition: 8. No VC money, no marketing spend.

Lovable, an AI app builder from Sweden, went from $0 to $200M ARR in 12 months, one of the fastest software growth curves in history.

Anysphere (Cursor) passed $1B in annualised revenue with fewer than 300 people. At $500M ARR, this worked out to $5M per employee, the highest revenue-per-employee ratio in the history of software.

These numbers are not anomalies. They are a signal. And if you ignore them because they look like Silicon Valley curiosities, you shouldn't, because the next threat to your market position will not come from your obvious competitor. It will come from a three-person team that did not exist a year ago.

The question is why. And, more importantly, what this means for your operating model.

AI is a multiplier, not an equaliser

The dominant narrative of the past two years has been that AI "democratises" productivity. That suddenly everyone is able to do things they could not do before.

That is not quite what is happening. What is happening is more interesting.

Take a person with no programming knowledge. With AI, they produce a value we can call 1. Without AI, they would have produced zero. The improvement is unlimited in percentage terms, but in absolute terms, the output is small.

An average programmer with AI: value 5.

A very good programmer with AI: value 10.

The same pattern applies to every function. The average product manager goes from 1 to 5. The good one goes from 5 to 10, and gets there much faster. The average designer reaches 5. The experienced one reaches 10.

AI does not deliver the same value to everyone. It is a multiplier of existing experience. The more experienced the user is in their domain, the bigger the multiplier.

On its own, this is not dramatic. The dramatic part happens when we add one more variable.

The 10×10 Rule

Think of someone who is both a programmer and understands design. They can translate design elements into code automatically, see trade-offs in real time, and make decisions that would otherwise require two people and three meetings.

With AI, this person does not become 10x better. They become 10 × 10 = 100x.

Why? Because cross-skills, when they live in the same mind, do not add up. They multiply. AI simply amplifies the multiplier across every dimension at once.

Add product or customer discovery skills to the same person. The product reaches x1000.

This is the 10×10 Rule. And it explains everything.

It explains why an 8-person team delivered $80M in value in six months. They were not working harder than a 500-person team. They had people with multiple skills in the same head, and AI was multiplying each of those skills at the same time. The founder of Base44 is the clearest example. He has publicly said that he has not written front-end code for months. AI writes the code for him. He then uses AI to make product decisions, uses AI to make design decisions, and uses AI to do customer discovery, all in the same session. Every one of his cross-skills is being multiplied by AI in parallel. That is how the output compounds.

It explains why Cursor reached $1B ARR with 300 people instead of 3000. They were not simply "more innovative". They hired a specific type of person: generalists with cross-functional fluency, for whom AI multiplies not one skill but three or four at once.

The Multiplier Gap

If the 10×10 Rule explains what is possible, the Multiplier Gap explains why most companies never get there.

You receive the mandate from the board to "adopt AI". You hire a Head of AI. You run pilots in support, in engineering, maybe in marketing. Each function sees a 2x or 3x improvement in some metric. You celebrate. You write a case study.

And then you wonder why the ROI never looked like the presentations you saw at Davos.

The answer is simple: you took AI as a sum instead of a product.

When you apply AI inside each silo separately, each function improves linearly. Engineering goes x10, but product cannot spec fast enough, sales cannot sell fast enough, legal cannot approve fast enough. The bottleneck moves, it does not disappear. One x10 is cancelled out by the next x1.

This is the Multiplier Gap. The distance between the multiplier you could have captured and the one you actually capture in practice. In theory x100, in practice x2.

That x2 is 2% of the value. And your 3-person competitor takes the other 98%.

What it looks like when it goes wrong: a familiar story

Let me describe a pattern I see repeatedly in companies trying to adopt AI "seriously".

The CEO announces an AI-first strategy. An AI Center of Excellence is set up. A budget is created for 10 to 15 AI use cases. Each function proposes one: a support chatbot, document generation in legal, lead scoring in sales, resume parsing in HR.

Twelve months later, six of the ten pilots have been frozen. The remaining four are running in production but with limited adoption. ROI is positive but mediocre. The CEO hears from consultants that "AI is overhyped" and starts losing interest.

Meanwhile, a competitor that started 18 months ago with three people now has twenty-five, and is taking your customers with products you are still discussing in committee.

The diagnosis is not that AI failed. It is that you applied it as if it were a software upgrade, when it is organisational redesign. You put faster tools into a slower structure, and were surprised that the speed did not change.

The question that remains is: what does a structure look like that is not slower than the tools running on top of it?

The organisational equivalent of AI-native teams

If an 8-person team can be worth $80M in six months, the goal for a larger company is not to shrink to 8 people. It is to artificially reproduce the conditions that enable this kind of speed.

These conditions have three components, and all three must be in place at the same time.

Cross-fluent teams, not just cross-functional. In an individual, skills live in the same mind and multiply. The organisational equivalent is not "I have a designer and an engineer in the same meeting". It is having people who actually speak the language of two or three functions. An engineer who understands customer discovery. A PM who reads code. A designer who understands unit economics. Without this, the team is just a co-location of specialists, not a product.

A shared context layer that works as a "shared brain". In an individual, whatever one skill knows, the other one knows automatically. In a company, this does not happen naturally. Context is scattered across dozens of tools, across Confluence, Slack, emails, ERP, CRM. To capture the multiplier, customer data, decisions, lessons learned from failures, product metrics, and market signals must be accessible to every function, and to every function's AI, at the same time. This is not a data warehouse project. It is a new architectural layer.

Short feedback loops. For an individual with cross-skills, feedback is instant. In a company, this translates into very short cycles between discovery, build, ship, and learn. If an idea travels from product to engineering to QA to marketing to sales over several months, the multiplication dies in the handoff delays. AI makes each function fast individually, but if the handoffs stay slow, you gain very little.

Taken together, the organisational equivalent is a synthetic entity that thinks like a person with multiple skills, with the same speed and the same coherence, but with the output of an entire company.

Why it rarely happens: the two levels of obstacles

In our conversations with a wide range of organisations, we see the same obstacles repeat. They fall into two categories.

Architectural obstacles relate to how information and work flow. Fragmented context: each function improves in isolation because its AI sees only one piece of the value chain. Bottleneck migration: when you apply AI only to one function, the bottleneck moves to the next one. Engineering produces x10, product cannot spec fast enough, sales cannot sell fast enough. The success of one function becomes frustration for the others. Speed mismatch: engineering and design adopt AI quickly, legal and compliance do not, and friction grows. These three are technical-organisational and can be solved with new architecture.

Human obstacles are harder. Middle management as a friction point: middle layers of management existed to coordinate and approve. When AI-enabled pods can self-coordinate, middle management becomes the point that cuts speed instead of adding it. This is the most politically painful part of the transition, and usually the one that gets ignored. Taste as the new scarce asset: AI amplifies judgement, it does not replace it. For decades, companies have been trained to hire for experience and specialisation, not for taste and judgement, and existing HR mechanisms do not know how to change direction. Wrong metrics: per-function KPIs show the wrong picture when value is produced in a multiplicative way through the pod.

The difficult part is that companies usually deal only with the first category. They buy platforms, unify data, run integrations. And then they wonder why the multiplier never kicked in, when the real bottlenecks were human.

The reframe that is needed

This is not a new problem. It is an old pattern in a new form.

When the industrial revolution brought mass steel production in the 19th century, civil engineers suddenly had a material with completely different properties from stone: lighter, more flexible, capable of much longer spans. And what did they do? For decades, they kept designing bridges the way they had designed them in stone. Heavy, massive, with short spans and huge piers. They were using the new material with the logic of the old one.

It took a generation for the Eiffels and Roeblings to appear and show that steel is not just stone that weighs less. It is a material that allows a completely different architecture: suspension bridges, cable-stayed designs, light structures with spans that would have been unthinkable in stone.

AI today is at the same stage. Most of the AI we see in companies is stone bridges made of steel. Same logic, same structure, just faster. The real multiplier only shows up when the architecture changes, not the material.

That is why the question is not "how do I add AI to my existing organisation". It is "what kind of organisation would I build if I started from scratch today with AI as a given".

The first frame gives you pilots that do not scale. The second requires a commitment to change four things at once: structure, metrics, hiring, and decision rights. Whoever pulls only one of the four levers gets linear improvement, and ends up believing that AI was overhyped.

AI adoption is not a project with a budget, a PM, and a deadline. It is an operating model change. And that is the central difference between companies that capture the multiplier and those that only see a linear improvement.

Four questions worth bringing to your next leadership meeting

If the change is structural, then the right questions are not about AI. They are about your structure. Here are four that are worth bringing to the next leadership meeting.

1. Take a typical decision you used to make five years ago, say a new product launch or a strategy shift. If you made it today with AI, would it have the same structure, just faster? Or would it actually be different as a process? If the answer is "the same but faster", you are building stone bridges out of steel.
2. If we were starting the company from scratch today with AI as a given, how would it be structured? Into how many pods, with what skills in each, and with how many approval layers between an idea and its execution? How far is that picture from our current structure?
3. Where we have applied AI, does each function improve on its own, or does the product between them kick in? Specifically: do our AIs share a common context, or does each one see only its own silo?
4. How many of our top 20 people actually have cross-functional fluency across two or three domains? If the answer is fewer than three, what is our strategy for the next 18 months?

The answers to these questions say more about your position in the next decade than any AI strategy deck.

Because in the end, the real question is not whether you will adopt AI. It is whether you will adopt it as a tool or as a structural element. Companies that see it as a tool will capture 2%. Those that see it as a structure will capture 100%. The difference between the two will not show up in quarterly results right away, but it will be visible in five years, when it will be too late to reverse.

The Multiplier Gap is one of the core patterns we observe in our conversations with organisations trying to do serious work with AI. If you recognised your own data in any of the above, the conversation continues here.

When people talk about AI activation, they usually mean the moment something gets automated. A workflow runs. A task is completed. A demo works.

That framing misses the point.

At helvia.ai, we do not think of activation as “the first automation.” We think of it as onboarding a digital employee.

What We Mean by “Digital Employee”

By digital employee we simply mean an AI agent built for a specific role, with clear responsibilities, boundaries, and access to the right company knowledge and systems.

A Digital Employee Has a First Day, Too

When we talk to a new client about building an AI agent, we deliberately use the analogy of hiring an employee.

Not because it sounds nice, but because it sets the right expectations.

Hiring an employee does not mean they become productive the moment they show up. You onboard them. You explain how the business works. You give them documentation. You define responsibilities, boundaries, and what “good work” looks like.

An AI agent is no different, with one critical distinction: You cannot talk to it.

You Cannot “Walk It Through” Things

A digital employee cannot ask clarifying questions. You cannot take it aside and explain something again. You cannot rely on shared assumptions.

The only way it understands how to do its job is through what you give it upfront.

That means written instructions describing the Job To Be Done, the individual steps of the job, supporting manuals, policies, FAQs, procedures, clear definitions of what is allowed and what is not, escalation rules, and concrete examples of correct outcomes.

It also means the content needs to be in the right format and structure to be digestible by an AI. If knowledge is scattered, inconsistent, or buried inside long documents that mix multiple topics, the agent will struggle to retrieve and apply it correctly. Clear structure directly impacts answer quality and reliability.

If something is unclear, missing, or contradictory, the agent will not raise its hand. The only way you notice is by observing how it behaves in real situations.

When it does something wrong, the root cause isn’t a “bad digital employee.” Usually, it is unclear instructions, missing documentation, conflicting rules, poorly structured content, or scenarios that were never written down because they were considered obvious.

The Hidden Work Is the Real Work

This is where most teams struggle.

Many business processes across industries are executed “by default.” They live in people’s heads. They are rarely documented because everyone assumes they are understood.

For a digital employee, nothing is obvious. If it is not written down somewhere, it does not exist.

In most organizations, the first blocker is not model quality, it is missing or implicit process knowledge. Our onboarding approach includes a content readiness pass, role and escalation design, and a supervised evaluation phase, so the agent earns trust before it scales.

That is why, when the material is missing, our first job is not to “train the AI,” but to help the client understand what needs to be created, what already exists, where the source of truth is, and how much effort it will take to make this usable, including bringing the content into an AI-friendly structure.

Only then can the agent be onboarded properly.

So When Does Activation Actually Happen?

From this perspective, activation is not the first automated task.

Activation happens when the digital employee has received its onboarding material, the instructions are clear and complete, the knowledge is structured in a way the AI can reliably use, and it has proven, in a real scenario, that it can perform its role correctly.

It is the moment when the user says:

“OK, I can delegate this.”

That is the first real day at work.

If you're considering an AI agent, we offer a 60-minute Digital Employee Readiness Check. Most companies discover they're not as ready as they thought, which saves months of false starts. You'll leave knowing exactly what you need to prepare before activation makes sense.

Get in touch at hello@helvia.ai to get started.

In today’s fast-paced business environment, organizations need HR solutions that are not only efficient but also intelligent and adaptable. Helvia People is an AI-powered HR platform designed to streamline HR processes, provide real-time insights, and enhance the employee experience—all while keeping people at the heart of every decision.

A Smarter Approach to HR

Traditional HR processes can be time-consuming, repetitive, and prone to error. Helvia People changes that by acting as an automated internal assistant that integrates seamlessly into Microsoft Teams. It leverages your company’s unique knowledge base to provide instant, accurate responses to employee questions, automating routine tasks and freeing HR teams to focus on what truly matters: supporting people and business growth.

Key Features That Make a Difference

• Intelligent Automation: From onboarding to daily HR queries, repetitive tasks are handled efficiently, reducing errors and saving valuable time.
• Real-Time Insights: Access up-to-date data on employee engagement, attendance, and other HR metrics to make informed decisions faster.
• Seamless Integration: Designed to work within Microsoft Teams, Helvia People meets employees where they already communicate and collaborate.
• Human-Centered Design: The platform prioritizes usability, ensuring that both employees and HR professionals can interact intuitively with the system.

Why Helvia People?

Helvia People empowers organizations to become more agile, productive, and employee-focused. By combining AI technology with a deep understanding of HR workflows, it ensures that every interaction is accurate, timely, and meaningful. The result is a more connected workforce and a smarter, more efficient HR function.

Looking Ahead

The future of work is digital, and HR must evolve to meet the demands of modern teams. Helvia People is more than just a tool—it’s a partner in building a workplace where employees feel supported and empowered.

Discover how Helvia People can transform your HR processes and enhance your employee experience today.

Learn More → https://people.helvia.ai/

This project is carried out within the framework of the National Recovery and Resilience Plan Greece 2.0, funded by the European Union – NextGenerationEU).

Helvia.ai was recently featured in Startupper MAG (#68), one of Greece’s leading entrepreneurship and innovation publications. In the interview, our co-founders Stavros Vassos and Dimitris Balaouras discussed the current state of Artificial Intelligence, the myths around an “AI bubble,” and why the real opportunity lies not in hype—but in digital labor powered by AI Agents.

Below, we’re sharing the key insights from that conversation, translated and adapted for our global audience.

AI Bubble or AI Boom?

The excitement around AI is undeniable—but so is the growing skepticism. Are we heading toward another tech bubble, or are we witnessing a genuine transformation?

According to Helvia.ai, the answer depends on which narrative you focus on.

There is indeed a level of hype, especially around AGI (Artificial General Intelligence)—the idea that machines will soon think and learn like humans. This narrative often comes with inflated promises that don’t reflect today’s technological reality. That’s where the perception of an “AI bubble” comes from.

However, there’s a second, far more grounded narrative: digital labor.

AI Agents are already operating in real production environments. They automate tasks, improve productivity, reduce operational costs, and integrate directly into business workflows. This is not a future vision—it’s happening now. This is the real AI boom.

For businesses—especially in markets like Greece—this shift represents a major opportunity. Companies that move beyond hype and adopt AI pragmatically can gain a significant competitive advantage, even against much larger organizations.

Why Companies Trust Helvia.ai

Many of Helvia.ai’s clients come with one of two challenges:

• They have specific, high-value use cases but lack the in-house expertise to implement them.
• They’ve already experimented with AI pilots that never made it to production.

What they find at Helvia.ai is not promises of “magic AI,” but real, production-ready digital labor.

Helvia works as a long-term partner, not just a technology vendor—co-designing solutions, implementing them, and continuously improving them based on real usage.

This approach is built on three core pillars:

1. Value-first solution design
   Clear scope, measurable KPIs, risk assessment, guardrails, and a realistic rollout plan from day one.
2. A production-grade AI Agent Platform
   The Helvia.ai Agent Platform includes built-in capabilities such as testing, versioning, logging, evaluation, integrations, and monitoring. This allows teams to focus on where value is created—process mapping and workflow integration—rather than rebuilding infrastructure for every project.
3. Clear boundaries and governance
   Agents are designed with explicit rules: when to respond, when to ask for clarification, and when to escalate to a human. This is critical in real business environments where operational risk matters.

All agents are continuously monitored in real conversations after go-live, allowing teams to identify failures early and apply targeted improvements.

What’s Next: Helvia.ai Agent Platform 6

Helvia.ai is currently finalizing Helvia.ai Agent Platform 6, scheduled to launch in early 2026.

This new version marks a major step forward in how AI Agents are designed, managed, and optimized, built around two core components:

• Designer - A powerful environment for building agentic workflows that combine multiple Large Language Model calls, enabling more intelligent, flexible, and robust agent behavior.
• Observatory - A comprehensive analytics and insights layer that gives organizations full visibility into agent performance and actionable, data-driven insights for continuous optimization.

With a strong focus on usability and UX, the platform gives organizations complete control and transparency over their AI ecosystem.

Why So Many AI Projects Never Reach Production

Despite widespread experimentation, many AI initiatives never make it beyond demos.

The most common reasons?

• Lack of a clear business use case
• Unrealistic expectations about current AI capabilities
• A tech-first mindset instead of a value-first approach
• Underestimating integration complexity with existing systems
• Ignoring organizational culture and adoption
• Treating AI Agents as “just prompts” rather than fully governed, monitored, and owned components of real workflows

Without proper governance, monitoring, and ownership, projects stall before delivering impact.

Practical Advice for Companies Starting with AI Agents

For organizations beginning their AI journey today, Helvia.ai offers clear guidance:

• Choose partners who know how to take AI to real production, not just demos.
• Start with clear, well-defined use cases and KPIs from the beginning.
• Move iteratively—from pilot to scale—rather than attempting risky, large leaps.
• Don’t be afraid to start simple. A small, focused agent can evolve into something much more powerful through real usage and continuous improvement.
• Invest early in data quality and production integration.
• Aim for small, consecutive wins, not one big bet.

The future belongs to the AI-native enterprise—and the encouraging news is that this future is more accessible than ever, even for small and mid-sized companies.

As Featured in Startupper MAG

This interview was originally published in Startupper MAG, Issue #68, highlighting Helvia.ai’s perspective on AI Agents, digital labor, and the practical path from experimentation to real business value.

We’d like to thank the Startupper MAG editorial team for the conversation and the opportunity to share our vision.

If you’d like to learn more about how Helvia.ai helps organizations turn AI into real digital labor, feel free to get in touch.

Found.ation and Helvia.ai join forces to help organizations embrace AI confidently and responsibly - ensuring that people remain at the center of transformation. Through this partnership, Found.ation brings its proven experience in business innovation and corporate training, while Helvia.ai with deep experience in Enterprise AI Agents contributes as the Program Tech Advisor. 

Across industries, one message is clear: AI is no longer optional. Yet while many organizations rush to adopt it, few manage to extract real business value. The challenge is not about accessing technology; it’s about understanding how to use it meaningfully. Most teams still lack a clear path for integrating AI into their workflows, bridging technical and human capabilities, and ensuring that new tools enhance rather than disrupt people’s work. 

This is the challenge that Found.ation and Helvia.ai decided to address together. 

In an era where technology and people must move forward together, this partnership reflects a broader shift toward integrating AI learning with organizational transformation. It bridges technological excellence with human-centered learning, combining Helvia’s expertise in Generative AI and language-AI with Found.ation’s experience in innovation and corporate training.  

Together, they aim to design and deliver a new generation of AI-powered business training programs that make AI accessible and actionable within business contexts, keeping human experience at the center. These programs focus on empowering people to make informed, ethical, and creative use of new technologies. 

Empowering teams to embrace AI with confidence 

This collaboration bridges two critical perspectives: Helvia’s technological leadership in Artificial Intelligence and automation, providing the technical foundation and technological tools, and Found.ation’s expertise in translating technology into learning and business transformation, ensuring that adoption aligns with each organization’s strategy and culture. 

Together, they translate complex AI concepts into practical learning experiences that help companies integrate new technologies effectively and responsibly. The programs are tailored to business needs, offered individually or even combined, supporting organizations in developing technical understanding and strategic readiness for the AI era. 

 AI-focused courses under Found.ation’s Business Training curriculum  

Within Found.ation’s Business Training curriculum, Helvia.ai acts as the Program Tech Advisor contributing its AI expertise and validating the technological accuracy and relevance of three key programs focused on empowering professionals to lead the AI era:  

• Transforming Business Using AI: A strategic overview of AI technology
  A masterclass for business leaders and decision-makers. This course provides a framework for understanding AI’s impact across industries. Participants learn to identify opportunities, mitigate risks, and align AI adoption with their organization’s goals and values.
• AI Prompting Essentials: Your first step to smarter GenAI use
  Participants learn the fundamentals of prompting, how to interact effectively with large language models, and how to integrate AI tools, such as Microsoft Copilot, into everyday workflows to enhance productivity, creativity, and insight. 
• AI Prompting Advanced: Unlocking the power of AI Agents
  Building on the foundations of the AI Prompting Essentials course, this program delves deeper into agentic AI, exploring how AI agents can plan, execute, and automate multi-step tasks. It equips participants to design workflows that integrate conversational and autonomous AI systems responsibly. 

Each program is designed by Found.ation in collaboration with Helvia.ai, combining business-oriented design with a human-centric learning approach. Each course carries the “Approved by Helvia.ai” seal, ensuring technical accuracy and alignment with real AI applications. The result is an educational experience that clarifies the role of AI and cultivates a culture of curiosity, adaptability, and ethical innovation. 

Bringing AI closer to business reality 

Through this partnership, the two organizations ensure that AI training goes beyond technical skills. Participants are guided to develop a strategic understanding of how AI augments human work, improves communication, and drives transformation. 

From understanding the difference between prompting and agentic workflows, to designing new roles for humans and machines in tandem, every course integrates hands-on learning, critical thinking, and real business scenarios that reflect real world needs. 

Leading AI integration with people 

While many organizations feel the pressure to “adopt AI fast,” Found.ation and Helvia.ai advocate for a more sustainable, human-centered approach. Their joint programs emphasize not only the “How” of using AI, but also the “Why” – ensuring that technology serves business purposes, human creativity, and long-term growth. 

This partnership reflects how educational design and AI innovation can come together to support organizations in navigating the AI era responsibly and confidently with a people-first approach. 

By Stavros Vassos, Co-Founder & CEO at helvia.ai

This post is based on a recent talk I gave at the 4th GenAI Summit on the 24th of November, 2025. You can find the slides of my presentation here and the recording of the full session here. My intention in the talk and this writeup is to share some thoughts on three things that I see people asking all the time about AI:

1. Is it an AI Bubble or AI Boom?
2. How can I benefit from AI Today?
3. What is a promising technical path forward?

First, I'd like to go through a quick overview of the last three years with GenAI.

Three Years of GenAI: A Rapid Transformation

The time of my talk (November 24, 2025) marked almost three years since ChatGPT first introduced Generative AI (GenAI) and Large Language Models (LLMs) to the world.

Key milestones include:

• November 30, 2022:  ChatGPT by OpenAI came out and shocked everyone with what was possible at that time! At the time it was using a version of GPT-3.5 that was called "InstructGPT".
• March 14, 2023: Then a few months later GPT-4 came out and shocked everyone again with more capabilities that were considered an order of magnitude more powerful! Then not other big shocks took place but lots of things happened.
• April 18, 2024: Llama 3’s by Meta release set a framework for open source (OSS) models enabling progress and growth in LLM providers.
• June 21, 2024: Almost 1.5 year ago Claude 3.5 is released by Anthropic, shortly becoming the "coding king," and essentially making coding a prominent use case for LLMs. This led all major LLM vendors to develop models specifically tailored for coding.
• November 2025: If we look at just the last couple of weeks, we see a new release of the latest version of LLMs by many major vendors (including GPT-5.1, Gemini 3, Claude Opus 4.5), as well as open source models (Kimi K2 Thinking, DeepSeek 3.2) that seem to be on par with the closed source models that are leading the race. 

In just three years, LLMs became the cornerstone of a new tech stack, and a rapidly evolving ecosystem that blends research and development of new applications and services.

AI Bubble or AI Boom? Two Visions of the Future

There is a wild spectrum of expectations about AI and there are passionate debates about whether AI works or not. Let’s look into two common perspectives:

• “Existential” Vision: AI will evolve to Artificial General Intelligence (AGI).
• “Industrial revolution” Vision: AI will bring us Digital labor.

1. The “Existential” Vision: AGI

According to this vision, AI will bring us Artificial General Intelligence - AGI. There are many definitions for this, let's go with the following for today. If you have an AI system that can perform equally or better than any human on things you can do while sitting on a computer, then we have AGI.

What AGI means

• Super-intelligence! Many people believe that since AI iterates faster on its capabilities than a person, ΑΙ bootstraps itself to unbounded skills.
• Which means that AGI essentially brings unbounded acceleration on science, economy, everything really.
• And to many it also means that there is a possible extinction of humanity.

Either way, whoever gets there first will have tremendous advantage! No wonder literally hundreds of billions of dollars are invested on the frontier labs pursuing larger LLMs and AGI.

Is AGI coming soon?

To be honest, I never thought nor had the feeling that this is going to happen soon. It's not an easy subject of course, and there is debate and strong opinions by very prominent figures in AI, e.g.:

• Geoffrey Hinton, widely known as one of the "godfathers of AI", has maintained many years now (one early reference can be found here) that we have just a few years to prepare ourselves before AI surpasses us.
• Yann LeCun, a high-profile AI researcher at Meta, has been arguing that LLMs are a dead-end and we need to start different lines of research (one recent reference here).

If they can't agree, how can you or me decide on this?!

Scaling toward AGI

One way to form an opinion is to see how things scale within the last few years. There is a quote from the CEO of OpenAI from almost a year ago that I think shows a common line of thought about how AI can grow to become AGI. It is from Sam's blog on February 10, 2025 and it goes more or less like this:

• The intelligence of an AI model roughly equals the log of the resources used to train and run it.
• These resources are chiefly training compute, data, and inference compute.
• It appears that you can spend arbitrary amounts of money and get continuous and predictable gains.

By the way, this is perhaps an optimistic view, especially coming from the CEO of a leading LLM company, and one whose funding relies on such assumptions about the future of AI. There is an important detail in this line of thought too: as mentioned in Sam's blog, the assumption is that the intelligence of an AI model roughly equals the log of the resources used to train and run it. Since log is not always easy to understand, let's see it with some examples of what this means (hand-wavy but true to the spirit of the assumption).

According to this:

• To "double” intelligence, you might need 10× the resources
• To "triple” it? Maybe 100× the resources
• To "quadruple" it? Maybe 1000× the resources
• This makes progress increasingly expensive and perhaps infeasible due to bounds on infrastructure or energy or.. earth size!

Overall, there is an indication that more resources will give us better AI and get us closer to AGI – but it could be 5 years or 500 years to reach AGI and may require other unconceived breakthroughs. Whether it's close or not seems to be more of a "gut feeling" and less of a predictable extrapolation of current results. So, let's say this is a "bubbly" expectation!

2. The “Industrial Revolution” Vision: Digital Labor

According to this vision, AI will bring us Digital labor, that is, digital employees that train themselves, onboard instantly, work 24/7, and cost less than human employees. This vision too is kind-of sci-fi coming to life, and at the same time there is lots of investment on supporting these new AI roles that will be working side by side or replacing human roles.

I would like to look into three prominent use cases where we see AI roles already being explored:

• Personal assistant
• Business automation
• Coding

AI roles - Personal assistants

Every “LLM app” like ChatGPT, Claude, Gemini, etc, acts essentially as a powerful personal assistant that operates on the digital world. Looking into some recent numbers about ChatGPT as reported by Financial Times:

• ChatGPT has more than 800 million regular users
• 5% of those are paying subscribers
• This amounts to 70% of OpenAI annual revenue

Judging from OpenAI alone, the personal assistant is probably the biggest use case of GenAI at the moment! And in fact, these AI personal assistants are already changing how we write and read text, how we search online, how we shop, and there is plenty of room to grow.

AI roles - Business automation

This is the core of the "Digital revolution" vision, according to which many human roles are going to be transformed into AI roles. Here we have mixed results. These are two articles coming from Gartner that are indicative of how this idea has evolved in the business world in the last couple of years:

• Gartner, Aug 2023: By 2026, investment in generative AI will lead to a 20% to 30% reduction of customer service human roles. (source)
• Gartner, Jun 2025: By 2027 50% of organizations will abandon plans to reduce customer service workforce due to AI. This shift comes as many companies struggle highlighting the complexities and challenges of transitioning to AI-driven customer service. (source)

The first article form a couple of years ago shows the initial excitement that in customer service we will have already replaced 20-30% of the human roles by today, while the second article of a few months ago reports on how this plan proves to be more difficult than initially thought. This is indicative of many cases trying to replace human roles entirely by so-called AI Agents. (Note: I got these two articles by a recent post of the AI Realist who does an excellent job fighting the hype!).

But what is an AI Agent exactly?

I often joke that an AI Agent is this mythical (AI) creature that is going to do all kinds of things and change our lives. There are plenty of definitions, most of them technical, e.g., a common one is that an AI Agent is an LLM-based system that is able to use so-called tools.

I like to use a more high-level definition that is more practical and helps understand what AI Agents represent today, and it goes like this:

Of course with LLMs we don't know exactly what they are going to write every time, there is this uncertainly. But with an AI Agent the uncertainty is about what they are going to do next: Are they going to send an email? Call an API? Send a message? Or something else?

AI Agents operate on an implicit tradeoff between these two desired qualities:

• Powerful – being able to handle scenarios that are not explicitly “programmed” for.
• Predictable – acting in the way an existing employee would handle things.

This is one of the major reasons why the promise of business using AI roles widely is not happening instantly. It is possible though, just not instant!

At Helvia.ai we have worked with large organizations and we have employed AI roles in business operations successfully and we have proved that you can get meaningful gains and ROI. The key is to do careful preparation and consider AI Agents in the scope of a wider Business and AI Transformation.

AI roles - Coding

Coding has become a very large niche and there is a long list of tools that apply to the whole software engineering lifecycle, some examples:

• New command-line tools (CLI): Codex, Gemini-CLI, GitHub Copilot, etc.
• New integrated development environments (IDE): Windsurf, Cline, Cursor, Antigravity, etc.
• New prototyping tools: Lovable, v0.dev, and others.

The high demand has led LLM providers create specialized models for coding and we see a release of something new every couple of weeks! In some sense, coding with AI is similar to business automation with AI, but the coding / software engineering environment is more structured and much faster to iterate, which makes it a great application domain for AI.

Results so far seem to show that AI can:

• Help junior software engineers learn faster if used as training aid.
• Assist senior engineers do the heavy lifting of projects.
• Change the way software engineers interact with information
  on the web.

Overall it provides the grounds for accelerating software engineering.

“AI Industrial Revolution” Vision – Are We There Yet?

So what's the verdict on this one?

What’s in it for You? How to navigate in this new world?

In practical terms, no bubble. In investment terms, it's another story. Let's say this in another way too. The technology is real: LLMs bring a new way of building "computers" and apps. It's the “Decade of AI Agents” in which we will transform culturally, socially, and operationally to incorporate new forms of autonomy in our personal and work lives. How can you make the most out of it now? I'd like to share some thoughts:

• At a personal level
• As an SME
• As an Enterprise

What’s in it for You As a Person

It's like all the leading labs are building flying cars and they give away free kits to try flying in your backyard! There is cutting-edge new technology being built every day and literally everyone can use the latest version of it as it comes out! In fact all major AI vendors offer a generous freemium and a ~$20 per month license for their personal assistant. This is a ticket to reinvent your interaction with technology, using AI as the ultimate interface to everything.

And how should you start? My personal favorite is if you don't know how to code, learn how to code! Or “vibe code” something for fun :) Here’s a mini game I did while creating this presentation, inspired by the bubble talk and an arcade game I used to play when I was young: Bubble Arcade by Helvia

What’s in it for You As an SME

As an SME typically there is no budget to build a tailored solution just for you, therefore you should not expect to get an AI Agent that solves exactly the problem you have in mind. However, you can use the built-in AI Agents on the Software-as-as-Service (SaaS) business applications you use, as every major SaaS provider is racing to add an in-app AI Agent on their offering. E.g., Atlassian offers Rovo to help with Jira, etc. Also, a more hands-on / DIY direction is to learn how to craft automation pipelines with AI-ready automation tools, e.g., use Make to setup email automations. There are hundreds of tools to use with affordable cost and great ROI to help you become the AI expert of your own SME.

What’s in it for You As a Large Organization

Let's first say that the AI Industrial Revolution is for you!

At this stage of technology and commercialization of LLMs, the AI Agents are most impactful when specialized on the actual business processes and requirements – and large organizations and enterprises are in the right position to make this happen as part of an AI transformation.

At Helvia.ai we have been working with enterprises almost 10 years now and we are happy to start a journey together to bring AI Agents in the places it makes more sense and brings more value – send us an email at contact@helvia.ai to get started if you are interested.

Looking Ahead: Beyond LLMs

Artificial Jagged Intelligence (AJI)

First, I would like to highlight something about the type of AI we have now. I like the term Artificial Jagged Intelligence (AJI) that Andrej Karpathy coined a little more than a year ago:

"The word I came up with to describe the (strange, unintuitive) fact that state of the art LLMs can both perform extremely impressive tasks (e.g. solve complex math problems) while simultaneously struggle with some very dumb problems."

There are plenty of examples of "silly" mistakes by LLMs, such as confusing that 9.11 is larger than 9.8 (which has to do with a way of interpreting 9.11 as a section number or a date, rather than a rational number).

My favorite one that you can try today has to do with visual models and generating images. If you ask a model to make an image transparent, most of the times what it will do is replace the background with a checkerboard pattern like this – which is something I actually tried earlier for this presentation on one of the latest and most powerful models.

I find this very interesting and it's a nice way to remind ourselves that the capabilities of the models today come to a great extend from reusing patterns they have "seen" at training time.

Why does this happen?

My guess is that most models struggle with generating transparent images because many images online that are marked as transparent are not available for direct download, and the training set did not include the actual transparent image to put as an example. Instead the training set probably included the preview of the transparent image online that looks like the real one but with a checkerboard background to highlight the transparency.

Neurosymbolic AI

And this brings me to a subject I am very fond of, namely, Neurosymbolic AI. As with many things with AI there are many definitions for this as well and it's an emerging field.

The easiest way to describe it is mixing methods and techniques from two disciplines in AI: the neural networks camp and the symbolic logic camp. I've been around in AI research enough to have seen an AI winter and an AI spring and to observe the shift of attention and power dynamics between these two camps. When I started my PhD the symbolic logic camp was ruling and the neural net camp was setting the stage for the dominating the field as it has done today.

It's interesting to observe these type of dynamics as it influences and sometimes defines the research that is visible to new researchers. For example now AI is almost (if not 100%) synonymous to neural networks and perhaps Generative AI. But AI is not just this, it's a complex network of disciplines that is not easy to navigate if you are not an expert. Notably, the AI literature includes a long list of research on symbolic logic, knowledge representation, and a different type of automated reasoning than the one we are used to see in the LLM space.

Now as the LLM hype is settling down, we see that it's not easy or immediate to solve everything with simply writing instructions via prompting, prompt engineering, context engineering, etc. Moreover, we see the prompts becoming more of a programming language with their own coding-like constructs such as conventions about <final_answer_formatting>, <user_update_immediacy>, <frequency_and_length>, and more (as per the latest OpenAI prompt guide), as well as coding-based specifications for tools and APIs. Overall, specifying what an AI Agent should do is getting more structured on top of the "just say what you think in natural language".

This type of structure has been the cornerstone of the other camp of AI, with symbolic logic being a strict/formal form of writing instructions for AI systems. I strongly believe that there is an untapped potential in combining tools and tricks from the symbolic logic camp of AI to the LLM space and the neural net camp of AI. There is work already at different levels of synergy: from all the way down into training different kinds of deep network layers to all the way up to having an LLM use a symbolic AI system as a "subroutine" or tool to solve a particular type of problem.

For a quick intro see this recent position paper in the upcoming AAAI conference: The Future Is Neuro-Symbolic: Where Has It Been, and Where Is It Going? I am excited by all the spectrum of directions being explored, and I believe that there are low-hanging fruits in a "service-like" synergy where an LLM consults with one or more expert symbolic AI systems to get help.

Lots of Opportunities for Service-like Synergy

As an example, consider the way we typically handle memory for AI Agents. AI Agents often need to keep a persistent memory as a blackboard of past interactions and notes that can help them provide contextualized responses based on what worked (or not) in the past.

This is typically done using LLMs to write the notes, a retrieval mechanism for getting relevant notes, and using LLMs to revise the memories. For simple use cases this works effectively, but as things get more complex then you need to refine and tune how the memories are stored, how they are retrieved, e.g., using some more elaborate retrieval approach with semantic search, and how they are updated in light of new information.

At the same time, there is a large literature on research on they symbolic logic side, on “knowledge representation” and “belief revision” that formalizes the flow of information, the rules, the defaults and exceptions, and other useful machinery to make bookkeeping robust, fast, and inexpensive!

Consider also that even if LLMs work well for managing memories in some cases, it is often not an optimal approach in terms of speed and cost. For example, setting the updated set of memories through an LLM requires in the typical case to generate the whole new updated set of memories. Similarly, keeping a representation of the state for a complex task faces similar challenges. Both cases are central in how AI Agents work and when complexity increases there are benefits in crafting an optimized approach. To this direction, there are lots of "deliberation algorithms" for drafting hypotheses and getting conclusions which can be leveraged as tools as part of the thinking process of an LLM-powered AI Agent.

Anyway, I probably got too technical for the scope of this post!

Closing Thoughts

Closing, let's reiterate on a single message that is worth taking away with you today. It's an exciting time to be alive technology-wise but also culturally and socially. We seem to be the ones that will get to shape how AI technology will be adopted and incorporated in our society, or at least set some pragmatic basis. As for AGI, the biggest frontier seems to be the physical world, i.e., how AI systems can actually live and work in the space we live and operate.

To me we will reach AGI when a robot can navigate in Athens and survive running errands in the city 😄

We are proud to announce that Helvia.ai has been awarded the Startupper Award 2025 in the AI category. This prestigious recognition highlights the innovative work our team has been doing in the field of artificial intelligence and underscores our commitment to driving technology forward.

The Startupper Awards, organized annually by Startupper.gr and Startupper MAG, celebrate startups and entrepreneurs who have made significant progress in their respective fields. Since 2016, the awards have recognized the creativity, resilience, and achievements of Greek startups.

The awards ceremony took place on Thursday, November 27, 2025, at the iconic Hellenic Cosmos, bringing together founders, investors, and industry leaders to honor the most dynamic and pioneering companies of the year. The event highlighted the transformative impact of Greek startups and demonstrated how they are shaping both the national and global technological landscape.

Being recognized in the AI category underscores the hard work, curiosity, and creativity of the Helvia.ai team. Every day, we strive to turn ambitious ideas into impactful solutions, and this award is a testament to that commitment.

We extend our sincere thanks to the StartupperGR team, our talented Helvia.ai team, and all of our supporters who have been part of this journey.

We are proud to announce that Helvia.ai is joining forces with our long-standing partner 01 Solutions Hellas, following their award of the national project to upgrade ERGANI, the Ministry of Labor’s core information system.

As a subcontractor in this landmark initiative, Helvia.ai will design and deliver the new Generative AI-powered Digital Assistant (AI Agent) for ERGANI—bringing advanced conversational AI into one of Greece’s most critical public digital platforms.

AI at the Heart of Everyday Employment Services

ERGANI plays a vital role in the daily functioning of the labor market, supporting:

• 2.2 million employees
• 600,000 businesses
• 5,500 public officials

Every day, the platform enables essential employment processes, including hiring, work schedules, overtime management, and leave declarations. With the introduction of an AI Agent, ERGANI takes a major step forward in accessibility, usability, and efficiency.

Our Mission: Simpler Access, Smarter Interactions

At Helvia.ai, our role is clear: to embed cutting-edge Generative AI and conversational technology into the daily interactions of millions of users.

The ERGANI AI Agent will:

• Simplify access to complex information
• Reduce friction in administrative processes
• Provide reliable, human-like assistance at scale

By transforming how users interact with complex labor processes, we aim to make public digital services more accessible, efficient, and human-centered.

Shaping the Future of Public Digital Services

Contributing to ERGANI’s evolution is both an honor and a responsibility. This project represents a meaningful step forward in the digital transformation of the labor market, and we’re proud to help shape the future of public-sector AI at national scale.

We look forward to helping shape the next generation of digital public services—where AI works quietly in the background, empowering better outcomes for everyone.

Artificial Intelligence (AI) often comes with a mix of excitement and inflated expectations. Is it a magical solution to every problem, or simply another tool in our digital toolkit?

In a recent radio conversation on ERT News, our co-founder Stavros Vassos addressed exactly that — exploring what AI really is, what responsibility in practice looks like, and how it will shape our everyday lives in the years to come.

👉 You can listen to the full conversation in Greek here (starts at 1h 3m 30s).

No Magic, Just Engineering

“There’s a bit of a sense that there are magical solutions” Stavros noted during the interview. “As in: ‘Oh, we have something difficult, let’s just have AI do it.’”

The truth is more grounded. AI is not magic. At its core, it’s software — systems built with data, rules, and feedback. These systems learn from prior decisions, but they don’t design themselves. Behind every AI stands a team of people who define its logic and outcomes. And that means responsibility lies not with the machine, but with its creators.

Responsibility in Practice

Stavros made it clear that responsibility and governance must guide AI’s development and application. “Ultimately, whoever builds the system then has the responsibility to make it objective.”

This insight reminds us that AI is not a shortcut — it requires careful design, ethical frameworks, and ongoing oversight to ensure fairness, transparency, and accountability.

AI in Everyday Life

One of the key points from the radio conversation was that AI will continue to blend into our daily lives, often working quietly in the background.

“I think that as the next years go by and AI is applied and tested more,” Stavros said, “we’ll see it everywhere in the facets of our daily life, not necessarily visibly, doing things that help us.”

From customer service interactions to smarter decision-making in organizations, AI’s presence may be subtle — but its benefits can be transformative when deployed responsibly.

Takeaway: A Future Built on Responsibility

The ERT News discussion underscored an essential truth: AI is not about magic. It’s about engineering, governance, and human expertise. At helvia.ai, we see responsibility not as an afterthought but as the foundation for deploying AI at scale — making it a force that enhances, rather than replaces, human judgment.

Artificial Intelligence (AI) is actively transforming the way businesses operate today. From streamlining customer interactions to enhancing internal workflows, AI-powered assistants are at the forefront of this change. In a recent interview on Naftemporiki TV, Stavros Vassos, Co-Founder and CEO of helvia.ai, shared insights on how the company is driving this transformation and what the future holds for AI in businesses.

Can AI assistants help in decision making?

The answer is yes, and it's all thanks to the new possibilities of artificial intelligence. While it may seem a bit daunting at first, the difference we see now is in the quality and extent of the assistance. In the past, a digital assistant could only give us a directed answer or guide us to a specific source of information. However, now it's becoming more flexible and can cover even non-standardized topics, all while answering in the way we want it to.

One practical way this technology is aiding businesses is by allowing us to track the progress of our company through graphs and numbers. We can ask our AI assistant to create a graph for us on the spot or even take a few steps further. The predictions of the future with AI may be uncertain, but the last few years have shown an increase in the efficiency of AI in the digital world.

Greek innovation on the global stage 

Greece is very optimistic about its tech talent! Companies such as helvia.ai have made significant strides in the global market, proving that Greek companies can compete with the best, even in fast-moving sectors such as AI. It's worth noting that although Greece isn't one of the first countries to embrace new technologies, both in the private and public sectors, the potential of people, especially in research and technology, is very high. This enables technology companies to grow, and attract international customers.

How to keep up with the constant evolution of AI

As a company implementing and developing AI solutions, helvia.ai finds the rapid pace of progress refreshing. To keep up with this pace, we are constantly integrating and bringing new things to the table. However, for companies looking to adopt AI, it can be a challenge. With the current hype and trend, it can be difficult to understand what is right for your business and even how to ask the right questions. Additionally, it's hard to keep up with the technical details that are vital when explaining the differences between solutions. It's like trying to find the diameter of a pipe when buying a car.

Do companies in Greece keep up with the pace of AI adoption?

The pandemic has accelerated digitalization and forced a second wave of digital transformation, including the adoption of AI. In the past, falling behind was a risk for not undergoing a digital transformation. Now, falling behind is a risk for not implementing new techniques that require a digital transformation. This acceleration is creating a differentiation between those who are keeping pace and those who are falling behind.

What the future of AI holds

Despite the rapid progress we have witnessed so far, we have not yet seen the full potential of these developments. The speed of progress has been so great that we have been unable to predict the extent of its impact.

While we cannot say for sure what the next few years will bring, we do know that there is still much to be done to fully integrate and utilize the existing advancements in AI. Therefore, even though we may not see developments as significant as those we have witnessed so far, we still have a long way to go to productize and facilitate the adoption of the existing powers of AI.

We’d like to thank Naftemporiki TV for hosting us once again. The full interview is available in Greek here

Whether creating content, translating text or even developing code, the best results from AI models depend on how well you formulate your prompts, your request. Creating effective prompts can to a great extent improve the quality and usefulness of the responses you will receive. 

Helvia.ai co-founders Dimi Balaouras and Stavros Vassos recently shared some tips in an interview on Flash.gr on how to write better prompts when using an AI model to perform various tasks.  

Prompting tip #1: Give  the AI ​​model a role or persona

AI models have been trained in a specific way, and giving them instructions that resemble what they have seen during their training helps them follow instructions more accurately. Even if you assign the model a role or persona that it has not encountered before, it will find clues to match it with familiar roles and choose how to proceed.    

Prompting tip #2: Provide clear and specific instructions

Provide detailed instructions that outline the general direction, as well as specific steps. Avoid using self-evident language because an AI model may not interpret it as intended.  

Prompting tip #3: Include examples  

Provide two or three examples of input and output to give the AI model a more tangible idea of what you want. This technique, called Few-shot, helps the model produce more accurate and targeted results by allowing it to compare and correlate the provided material with what it knows.  

Prompting tip #4: Use punctuation marks

The concept is to make the prompt as readable as possible. The instructions should be presented as small bits and pieces, which are easy to pick out. If we write the instructions as we would say them, some information may be lost.  

When for example we use ChatGPT, we use plain text because there is no rich text editor to add formatting. To capture formatting, we use a simple formatting called Markdown. 

When giving a series of instructions, it helps the AI ​​model to divide the concepts into smaller units, separated by numbering and blank lines.  

Prompting tip #5: Use Imperative language

AI models follow instructions to a degree, depending on how clear they are, but not always 100% of the time. To make an AI model follow an instruction more systematically, use absolute language and "order" it to not deviate.  

Prompting tip #6: Include your own sources as bibliography

When asking an AI model for information, such as "what is the population of Greece?", it will answer based on what it has stored during training, or it will do a Google search and try to answer. If we need to rely on our own data, we must add our own sources either by uploading files or by writing along with the instructions.     

This usually requires additional tools and techniques to work well, such as the Helvia.ai platform that allows adding a knowledge base, e.g. from the customer service department and getting a digital assistant that responds based on that information alone.    

Prompting tip #7: Ask the AI model to think in logical steps (step by step)

This technique prompts the AI ​​model to write down its thoughts as it processes the query and produces the answer. It makes the AI ​​model's process of arriving at an answer more understandable to the user and enhances the accuracy in taking logical steps. As it writes them, it also uses them as a guide for continuation, almost as if it is "writing on the board" as it thinks.    

Prompting tip #8: Outline the steps the AI model needs to follow to answer correctly

Even if the AI model suggested the steps required to come up with the response, writing them down directs it to get closer to what we want. 

Prompting tip #9: Use the AI model one step at a time

When it comes to performing multiple tasks, it's best to break them into smaller pieces and ask the AI model to perform each one separately.

Example:

• First we ask the AI model to extract the details of the conversation e.g. the project the conversation refers to, the project owner and the subject of the email.
• Then we ask the AI model to make a summary of the email and translate it into English.
• Finally, we ask the model, using the project details and the summary from the previous two steps, to write an email in English to send to a colleague who does not speak Greek.

If we didn't break it down into steps and asked the AI model to do it directly, the quality of the answer would be limited compared to doing it in steps.  

By following these prompting tips, you can enhance the accuracy and quality of the AI model's responses. Giving the AI model a role, providing clear instructions, including examples, using punctuation marks and imperative language, providing sources, thinking in logical steps, outlining required steps, and breaking tasks into smaller pieces are all essential components to get the most accurate and high-quality responses from AI models. Implement these tips and see the difference in the AI model's performance.

The full article in Greek  is available on Flash.gr.

If you have any questions or you would like to schedule a demo to see the helvia.ai conversational AI platform in action, contact us at hello@helvia.ai

Dimi Balaouras is the CTO and co-founder of helvia.ai and leads the technological innovation and product development of the company.

Stavros Vassos is the CEO and co-founder of helvia.ai. He is an expert in Artificial Intelligence and in particular in the fields of Conversational AI and Agentic AI.  

Online retailers are always looking for ways to improve customer experience, and one of the ways they're doing this is by using AI shopping assistants. However, with so much information to sift through, these assistants can struggle to provide accurate and personalized recommendations using RAG. That's where Hypothetical Document Embeddings (HyDE) come in. By leveraging these technologies, retailers can improve the performance of their shopping assistants and provide customers with a more efficient and personalized shopping experience.

At helvia.ai, we recently engaged in a comprehensive discussion about the potential functionalities of a GenAI shopping assistant, focusing on how it could accurately identify the products a user desires or needs for specific tasks. This discussion provided the perfect setting to initiate an exercise I call "Product the Gap", which was designed to explore consumer needs ("Jobs") and possible solutions using the Jobs To Be Done (JTBD) framework. The insights gathered were visually represented on a Miro board.

In this blog post, we'll go through our approach using some real-life examples.

How to Leverage an AI Shopping Assistant with RAG and HyDE?

The Jobs To Be Done (JTBD) framework

The process began by addressing the primary question: What is the main job of someone looking to purchase a product online? From this, we formulated the core job statement (Main JTBD (1.0)): "Enable me to efficiently discover and select products that align perfectly with my specific goals and tasks, minimizing both effort and uncertainty throughout the decision-making process." This can be viewed on the Miro board, marked with a red sticker No. 1.0

To better understand the job context, we documented several job examples on the Miro board, each tagged with a red sticker (No. 2.x). Here, we will discuss The Bluetooth Example (2.2), which closely aligns with our solution:

"I want to buy a Bluetooth speaker for under $100 to use at pool parties."

Exploring how a user might currently find and purchase the best Bluetooth speaker under $100 for pool parties involves the following steps:

• Search Google for articles reviewing Bluetooth speakers.
• Identify waterproof models.
• Conduct a feature and price comparison.
• Decide on one.
• Research where it can be bought at the lowest price.
• Make the purchase.
• Alternatively, the user might:
• Ask friends for recommendations.
• Search for model reviews on Google.
• Compare prices.
• Complete the purchase.

Simplifying the JTBD process with AI

Our solution simplifies this process: by requesting "three Bluetooth speakers for use at pool parties under $100," our AI will offer three options from our e-shop, complete with brief descriptions and pricing. You can then select one, add it directly to the shopping cart, or visit the product page. The response from this request can be seen under sticker 2.2.2 on our board at perlexity.ai.

Identifying the Product Gap (2.2.3)
We compared our AI solution with traditional methods to understand where our approach improves functionality:

(+) Reduces time spent researching.
(+) Quicker in finding and selecting products.
(-) Uncertainty about whether the best option was chosen.
(-) Traditional preference for Google searches over querying bots.
(-) Higher trust in personal recommendations over bots or Google.
(+) More suitable for advanced users familiar with AI, like ChatGPT.

This analysis raised an important question: "How can we integrate a search form as the start of a chat conversation?" We plan to explore this further in our design phase.

Repeating the Exercise
It is beneficial to repeat this exercise with different examples in varied contexts to fully understand the constraints and potential of our solution across different scenarios.

How It Could Work (3.1)
Our solution relies on an AI Assistant metaphor that is able to search over the product catalogue in a more effective way, typically through a Retrieval Augmented Generation (RAG) pipeline. However, creating an appropriate knowledge base for RAG using e-shop product data presents a challenge: the product descriptions may not explicit mention references to the product's suitability for specific uses like pool parties..

One idea is to create Hypothetical Documents using a language model to generate additional information about products, such as writing product reviews, or describing ideal usage scenarios or even describe the image of the product . This enriched content could significantly enhance the knowledge base, enabling it to accurately fulfill queries like "suggest three Bluetooth speakers for use at pool parties under $100." This is actually supported by recent research on what is called "Hypothetical documents Embeddings" which shows improvements in similar scenarios.

What are Hypothetical Document Embeddings (HyDE)?

Let’s imagine an e-shop planning to introduce a new Bluetooth speaker specifically designed for pool parties. This speaker boasts features such as high water resistance, extended battery life, and powerful bass suitable for outdoor environments.

To aid the e-shop’s recommendation engine in accurately categorizing and suggesting this new product, hypothetical document embeddings can be created. The system would generate an imaginary product description incorporating key features of both pool party-friendly equipment and robust Bluetooth speakers. For example, it might include elements from existing durable, waterproof speakers that are known for high sound quality outdoors.

This hypothetical description helps the recommendation system understand how this new speaker aligns with potential customer interests who often search for pool party gear or outdoor sound systems. By training on this constructed scenario, the system can more effectively suggest the speaker to those looking to enhance their pool party experiences, ensuring targeted recommendations even before actual customer feedback is gathered.

Testing Hypothetical Documents

To evaluate the practical application of hypothetical documents, we used them with three real products:

Test 1: IKEA HEJNE 3 Shelves (3.2.1)

We generated the following hypothetical documents using ChatGPT-4 with these prompts:

• “Write a review for IKEA HEJNE 3 shelves”
• “Best use of HEJNE 3 shelves”

These hypothetical documents could facilitate queries such as "I want shelves for my garage" helping to efficiently retrieve the related product linked to the hypothetical document, which would not be possible to retrieve using just a RAG.

View the Hypothetical Documents on the Miro board, marked with a red sticker (No. 3.2.1).

Test2: Amazon T-Shirt (3.2.2)

We generated the following hypothetical documents using ChatGPT-4 with these prompts:

• “You are a stylist, can you describe the tshirt, what style it is and what other clothes I can match it with?”
• “You are a stylist, can you describe in the image what she is wearing, what style it is and what other clothes I can match it with?”

The hypothetical documents supported queries such as "I want a retro-style T-shirt to wear with a denim jacket."

Test3: Vacuum Cleaner (3.2.3)

We generated the following hypothetical documents using ChatGPT-4 with these prompts:

• “Can you write a review for Dreametech T30 Cordless Vacuum Cleaner, 90mins Long Runtime Stick Vacuum, 190 AW Robust Suction Handheld Vacuum, Cordless Vacuum with HEPA Filters for Hard Floor Stairs”
• “For what kind of use Dreametech T30 Cordless Vacuum Cleaner is best for?

The hypothetical documents assisted with queries like "I want a vacuum suitable for pet owners that is easy to store."

These examples showcase the potential of leveraging Retrieval Augmented Generation (RAG) systems with hypothetical documents to effectively bridge the gap between complex consumer queries and relevant product suggestions.

Google it vs Chat it 

During our exploration, two other significant insights emerged:

1. They typically think about which keywords will yield the best results for their desired outcome, rather than simply asking for what they want.
2. Users have been conditioned to use search engines rather than chatbots. 

How can we determine the specific purpose a user has in mind when searching for a product?

Building on the second point, it might be interesting to reverse the Jobs to be Done approach. For instance, when someone uses a search field, we could prompt the language model to identify potential jobs that need to be done. For example, if someone searches for “Household Vacuum,” we could use the LLM to ask, “Give me only the Category/Context of the top 10 JTBD for users looking to buy a vacuum, based on popularity.” Tagged with a red sticker (No. 4.1)

The response of ChatGPT4:

Certainly! Here are the Category/Contexts for 10 different JTBD scenarios for users looking to buy a vacuum:

Efficient Household Maintenance
Allergy and Health Management
Family and Child-Friendly Homekeeping
Flooring Versatility and Care
Pet Ownership and Care
Professional and Commercial Cleanliness
Space Optimization for Compact Living
Accessibility and Ease of Use
Sustainable and Eco-Conscious Living
Vehicle and Furniture Maintenance

Now, we can use this list to ask the user “What is the use you want the vacuum for?“ and then use the Hypothetical documents to provide product results.

How might we integrate the user search flow with AI-driven conversational flows?

Implementing a prompt beneath the search box that asks, “What is the use you want the vacuum for?” could serve as a method to integrate the user search flow with AI conversational flows.

However, I am confident that there is ample room to explore more effective solutions.

Taking Your Shopping Assistant to the Next Level with RAG and HyDE

As we wrap up this exploration into the evolving role of AI in shaping the future of online shopping, it’s clear that hypothetical documents and advanced retrieval systems like RAG can significantly enhance the way we search and discover products online. Our experiments with the "Product the Gap" initiative and hypothetical documents have opened up new pathways to connect complex consumer needs with precise product suggestions.

If you are interested in applying this AI-driven approach to enhance your own products, we would love to collaborate. Contact us to explore how these innovative technologies can be tailored to fit your business needs and elevate your customer experience.

Together, let’s continue to push the boundaries of what GenAI can achieve in the world of online shopping, making it more intuitive, efficient, and aligned with user expectations.

Helvia.ai’s CEO, Stavros Vassos, participated at the event "The Digital Era, Artificial Intelligence and Techno-Ethics", hosted by the Association of Industries of Thessaly & Central Greece (AITCG), on 23 February 2024.

The aim of the event was to present experiences and perspectives on how technology is shaping our world and the challenges and opportunities it brings along. Recognizing the importance of digital transformation for businesses, the economy, and society, the AITCG has set up a special working group within the framework of the "Act & Grow 25" program. This group focuses on addressing digital transformation issues and aims to identify the needs, priorities, and concerns of the Association's member companies regarding their digital transformation.  

The event featured distinguished speakers, including the Minister of Digital Governance Mr. Dimitri Papastergiou, who shared their expertise and valuable experiences on topics related to the digital age, artificial intelligence, and techno-ethics. Stavros Vassos was among the speakers who discussed various aspects of digital transformation and AI.  

Dr Vassos highlighted the recent waves of digital transformation, comparing the emergence of ChatGTP with the earlier impact the pandemic had in the quick adoption of digital tools. He also acknowledged the desire of businesses to integrate new technologies, such as AI but noted challenges such as the need for infrastructure and the time-consuming process of building systems. Dr Vassos also discussed the evolution of AI from basic systems to sophisticated ones, capable of solving a variety of problems in different domains, at a much faster pace, almost in a zero-setup manner.

Additionally, Dr Vassos shared helvia.ai's experience, covering areas such as automating conversations, improving the customer and employee experience, and end-user engagement. He also noted the increasing acceptance and adoption of digital technologies by organizations, highlighting the trend towards closer engagement and continuous evolution in the digital area.  

In summary, the event provided a platform for experts to share their experiences and insights on how technology is shaping our world. The discussions on digital transformation, AI, and techno-ethics highlighted the challenges and opportunities that lie ahead for businesses, society, and individuals. As we continue to navigate the digital age, it is crucial to keep abreast of these developments and harness the power of technology to drive growth and progress.

The interview is available in Greek here and you can see some highlights from the event here.

We would like to thank the Association of Thessalian Industries for the invitation and the opportunity to attend the event.

In a recent episode of Monokeros (which means unicorn in Greeκ) dedicated to AI, Stavros Vassos, the CEO and Co-Founder of helvia.ai discussed with Giannis Dionatos the company's origin, its clients, and the rapid evolution of artificial intelligence.  

Helvia.ai: helping businesses transform with AI 

Artificial intelligence (AI) has become ubiquitous in our daily lives, and its capabilities are evolving, allowing us to solve problems more efficiently and effectively. One such application of AI is in the development of chatbots, which are becoming increasingly popular in the business world.  

Founded in 2016, helvia.ai specializes in creating AI chatbots that equip companies with communication solutions. Helvia.ai’s no-code platform allows businesses to build automated chat within their systems, making communication easier for all parties involved.  

AI Chatbots: the future of communication   

AI chatbots are systems that offer dialogues, making communication easier for customers and employees alike. They enable communication, even in simple matters such as the location of a package that has been ordered, or even more complicated, step-by-step guidance. With a wide range of applications, AI chatbots are becoming increasingly popular in the business world.  

In recent years, chatbots have become popular in two main areas of business. The first is customer service, where chatbots help end-customers learn and solve their problems. The second is internal problem-solving, where chatbots help employees and partners with their work. Helvia.ai's platform is centered around making these solutions customizable to meet the specific needs of each organization.      

Helvia.ai's journey: from Greece to the US    

Starting a technology start-up in Greece may be challenging, but both co-founders had spent many years abroad and were familiar with the communication channels used in North America. This knowledge gave them an advantage in finding clients from the other side of the Atlantic.  

In 2017, helvia.ai managed to find large clients outside of Greece who trusted them and are still with them. Their success with big clients in the US shaped the way their platform works, making it easier to address the needs of larger organizations.     

Helvia.ai’s focus in Research & Development

Artificial intelligence (AI) has come a long way since its inception and is now playing a significant role in many aspects of our lives. In business, AI is being used to improve efficiency, reduce costs, and provide better customer service. Helvia.ai is at the forefront of implementing new AI technologies and bringing innovative products to the market. In addition to drawing on the results of research and development from universities and the wider community, helvia.ai also has its own research department to create new tools and solutions, gaining a competitive edge in the market.     

Helvia.ai’s platform has already integrated functional, generative AI and has launched solutions both in Greece and abroad. In addition to that and as the greatest intensity of research globally is focused on developing new large language models (LLMs), helvia.ai is developing a hybrid approach for employing LLMs that reduces costs and improves efficiency. This paradigm combines specialized smaller models with the power of large language models, making their use efficient and cost-effective.

Looking into the future of Gen AI 

The potential uses for AI are vast and varied. In healthcare, AI can help diagnose diseases and assist with treatment plans. In education, AI can personalize learning and provide more effective feedback to students. In transportation, AI can improve safety and efficiency. As AI becomes more advanced, its uses will continue to expand.     

Helvia.ai is committed to staying at the forefront of AI and developing innovative solutions that not only improve business efficiency but also benefit society as a whole. As we continue to explore the potential uses for AI, it is crucial to consider the ethical implications and work towards solutions that are beneficial to all.

The interview is available in Greek here:

Contact us at hello@helvia.ai to discuss your conversational GenAI needs.

Helvia.ai is excited to be partnering with Novibet, a leading GameTech company, for its initiative to enhance customer experience with AI chatbots. The GameTech company, now present in 12 countries, is transforming customer service in Greek online gaming by focusing on AI customer support chatbots. This is an important milestone for the Greek tech community, with the new AI assistants using natural language processing and deep learning to provide customers with fast and accurate answers to their queries.

Novibet's commitment to effective customer support aligns perfectly with helvia.ai's mission to enhance customer experience with the latest AI technology. Helvia.ai’s AI assistants use natural language processing and deep learning to provide Novibet customers with fast and accurate answers to their queries, ensuring that they have the best possible experience. 

Dimitris Balaouras, CTO of helvia.ai, expressed his excitement about the partnership, saying, "Working with one of the biggest players in the industry is a great honor and we are confident that our AI assistants will offer Novibet customers a new experience, setting a new standard in the space.”   

Aphrodite Pina, Director of Customer Operations at Novibet, highlighted the importance of keeping the human factor at the core of the customer experience while leveraging cutting-edge technology. She said, "As we delve into the field of artificial intelligence, the integration of AI assistants promises to revolutionize the way our customers interact with our platform by providing a personalized gaming experience free from time-consuming processes. We believe this partnership will not only upgrade our services but also set a new benchmark for the entire industry, staying true to our commitment to provide the best customer experience."   

At helvia.ai, we are thrilled to be part of this project, which is the first large-scale Gen AI project for Customer Experience in Greek. We are confident that our AI assistants will help Novibet provide their customers with a new level of service, setting a new standard in the online gaming industry.   

This partnership is an exciting development for the Greek tech community and the online gaming industry as a whole. Helvia.ai is proud to be a part of this innovative project and working with Novibet to transform customer service in Greek online gaming.