From ELIZA to AGI: A Personal History of Chatbots - and Why the Future Finally Caught Up

Now, hiring. Customer Journey Specialist. Humans Need Not Apply. 

In October 2016, I presented a keynote to Tamedia Digital, the digital arm of the traditional media company based in Zürich, titled “Bots, Bots, Bots - Hype or the Future?”.  This talk happened only a few days after my own identical baby twins were born, so it felt all the more relevant to be discussing the future of bots and digital twins. 

Click on the image to view the Tamedia keynote presentation.

At the time, chatbots were experiencing their first mainstream boom. Facebook had just opened its Messenger Platform, “conversational commerce” was the phrase of the moment, and everyone from Icelandic Air to 1-800-Flowers was experimenting with scripted chat flows that felt - depending on the hour - either magical or maddening.

In parallel, at my company Fractal Labs (now trading as tomato pay) we had just launched one of the earliest automated financial assistants for small businesses: a mobile-first, conversational interface designed to help entrepreneurs understand and manage their finances in real time. For us, this wasn’t about novelty. It was about reducing the cognitive load for business owners - a psychological problem as much as a technical one. My education in psychology had taught me how humans process information, respond to cues, and build trust. My years at Google had drilled into me the value of scale, data-driven decision-making, and user-centred design. Fractal was my attempt to bring these disciplines together.

Now, nine years later, with the rise of agentic AI, GPT-style language models, and autonomous workflows, the entire conversation has shifted. But to understand why chatbots finally work today - and why 2016 was only the prelude - we need to look back.

A Brief History of Teaching Machines to Talk

In my Tamedia keynote, I shared a timeline beginning with ENIAC in 1946 and stretching through the birth of linguistics, early AI experiments and modern digital assistants. Even then, the through-line was clear: the story of chatbots is the story of our ambition to create interfaces that feel natural, intuitive, and human.

Here are a few examples worth highlighting:

• Alan Turing (1950) gave us the philosophical challenge: can a machine think?

• Noam Chomsky (1957) provided the linguistic scaffolding: how do humans generate meaning?

• ELIZA (1966) proved that with clever pattern matching, machines could appear empathetic.

• SmarterChild (2000) brought bots into the living rooms of teenagers on AIM and MSN.

• Siri (2011), Google Now (2012), Alexa (2015) signalled the arrival of ambient computing - assistants that lived in our pockets and on our kitchen counters.

However, these systems were brittle, scripted and limited. They were performers reading lines, not actors improvising.

Even in 2016, I noted: “Bots are not new. We simply have more processing power, more data, and more behavioural context than ever before.” The foundations were there - but something was missing.

Why 2016 Was a Turning Point (But Not the Destination)

When the Messenger Platform launched, everyone believed messaging would replace apps, and in some ways, it did. Users were already spending half their mobile time inside just a handful of apps; mostly WeChat, WhatsApp, Instagram, and Messenger.

What we learned during those years was essential:

• Simplicity: users want ease and convenience.

• Natural: they want to speak in their own language - not the language of rigid decision trees.

• Content: most importantly context is king.

The technology simply wasn’t ready. Bots couldn’t truly understand intent, personalisation was shallow, and as we saw with Microsoft’s infamous Tay experiment, poorly trained models could spiral in unpredictable - and often toxic - directions.

While Fractal’s financial assistant was innovative for its time, it still hit the ceiling of 2016-era NLP. Getting a chatbot to understand that “What’s my cash flow looking like next Friday?” and “Am I gonna run out of money soon?” were the same questions required heavy engineering and a lot of hand-holding.

We had the vision but the models weren’t yet capable of matching it.

The Leap From Chatbots to Agentic Intelligence

Fast forward to today. The shift isn’t merely another generational upgrade - it’s structural.

Large language models don’t just process language; they generate reasoning, inference, planning and action. They understand nuance. They adapt. They can orchestrate multi-step workflows. We’ve moved from bots that respond to prompts to agents that can autonomously pursue goals.

This is the future that early chatbot pioneers imagined but couldn’t reach.

If the 2010s were the decade of conversational interfaces, then the 2020s are the decade of agentic AI - systems that can actually perform tasks, not just chat about them.

This evolution is deeply personal to me. The same mission that guided Fractal - reducing cognitive load for entrepreneurs, automating the mundane so people can focus on what matters - now underpins the broader work I’m doing to build an AGI-driven accounting automation. The dream of 2016 can finally be executed with the technology of 2025.

Looking Forward: The Human/AI Co-Pilot Era

If psychology has taught me anything, it’s that technology succeeds when it respects human limits and amplifies human strengths.

The most powerful applications of AI emerging today - in finance, accounting, commerce, and beyond - are not about replacing people. They’re about designing systems that understand behaviour, anticipate needs, and quietly lift the cognitive burden.

Back in 2016 I ended my keynote with three design principles that feel more relevant than ever:

1. Stay in context.

2. Keep it simple.

3. Listen, learn, iterate.

The tools have changed. The stakes are higher. But the principles remain the same.

We are finally stepping into the world early AI promised - one where machines don’t just talk, but help.

In a way, the automated financial assistant we built at Fractal was a prototype of this future: an early glimpse of a world where your business has a co-pilot, your finances have an interpreter, and your workflow has an intelligence of its own.

The difference today is that the technology finally matches the ambition.