From chatbot to AI employee. Ten years, five shapes.

Jack and I started Rebotify in Melbourne in July 2016. Ten years this winter.

In that decade we have built, more or less, five different things and called them all "the product." Every one of them was, in hindsight, a step toward a question we did not yet know how to ask. The question, we now think, is the only one that matters: what outcome is the AI accountable for?

I want to walk through the five shapes, because the arc says something useful — to us, and I think to anyone trying to make AI useful inside a real business.

2016 — The rule-based chatbot.

We launched Rebotify as a no-code builder for chatbots on Facebook Messenger, Slack, WeChat and the web. Decision trees with a thin layer of intent classification on top. You drew the conversation flow, you trained it on twenty utterances, and the bot replied. We thought we were selling automation. We were really selling the appearance of conversation. A customer could ask the bot a question, and a scripted reply would come back. That was the whole loop. When the script ran out, the bot said "I’ll get a human" and the work landed back on a person’s desk anyway. The bot took the easy part. The team still owned the hard part.

We targeted small businesses, then pivoted to enterprise in 2017 after the Melbourne Accelerator Program. AGL, Tiffany & Co, BUPA, Open Universities. Real contracts. Real volume. And — quietly — a real ceiling. The bots answered. They did not do.

2018–2019 — API connections.

The honest fix was obvious. Wire the bot to the systems where the work actually lived. CRM, ticketing, calendars, internal tools. We added API connectors so a chatbot could create a ticket, look up an account, book a meeting, escalate to the right team. Now the bot was not just talking. It was clicking the buttons a human would have clicked, on the systems a human would have used.

This was the first version that ever felt like work was getting done instead of deflected. But the bot was still doing exactly what we told it to do, in the exact order we told it. Every new edge case was a new branch. Every new system was a new connector we hand-built. The team’s job had shifted from training intents to maintaining integrations. Less talking, more plumbing. Still the same conversation: who owns the outcome when the bot misses a step?

2022–2023 — LLM chatbots.

ChatGPT changed what the conversation layer could do, overnight. We swapped a lot of the intent-classification scaffolding for LLMs. Suddenly the bot understood phrasing the rule-based version would have routed to a human. Tone. Nuance. Ambiguity. The script burden dropped by an order of magnitude.

But — and this is the part I think the category mostly missed — an LLM bot is still a bot. It is still a thing the customer talks to, on the customer’s clock, about the customer’s question. It is faster to build than a decision tree. It is no more accountable for the work than a decision tree was. A great LLM reply that does not resolve the customer’s problem is still a deflection. We had made the conversation better. We had not yet made the work better.

2024–2025 — AI agents.

The agent wave was the first time the technology let us close the loop. Not just understand the question, not just call one API — actually plan, take a sequence of actions, recover from failure, and come back with a result. We rebuilt the core to run agentic workflows: research a lead, draft the reply, log the deal, send the report, watch the inbox, follow up tomorrow.

This is when we started using the word "employee" internally. Because the agent was no longer doing one trick. It was holding a small piece of someone’s role.

And this is also when we hit the next ceiling, the one that finally told us where we had been wrong all ten years.

The unit was wrong the whole time.

Every version of the product, all the way back to 2016, was sold and measured on the technology inside it. Number of bots. Number of integrations. Number of agents. Tokens. Workflows. Coverage. The slide always had a count on it.

But no operator we have ever met cares about the count. They care about whether a thing got done. Whether the reply went out before the customer churned. Whether the deal moved. Whether Monday morning is calmer than last Monday. The number of bots doing the work is the vendor’s problem. The customer’s problem is the work.

Once we wrote that down — once we said the unit of account is not the agent, it is the outcome — every other decision rearranged itself. Pricing stopped being per-token and became per-task, per-month, or per-result. The deliverable stopped being a dashboard and became a named role. The success metric stopped being "agent uptime" and became "did the work land, on time, at quality, with the customer’s tone."

This is also where we disagree with the recent wave of research arguing AI agents should not be treated as employees. The objection is right about the technology — an agent is an instrument. The objection misses the buyer side — the unit of account a business can scope, price and review is a role, not an instrument. We treat the employee framing as a buyer-side contract, not a claim about the model.

2026 — AI employees.

This is what we are doing now, and it is the first version of Rebotify I would call genuinely different rather than incrementally better than the last. One name in your inbox, owning one job. A role that exists because there is a piece of business that needs to land — not a capability, not a feature. A vault of memory the customer owns that compounds while the model under the hood changes. The shape of an employee, not a tool.

Why this is the light at the end.

For ten years we kept building more capable AI. Every shape was more capable than the last. Decision trees, APIs, LLMs, agents, employees. And for nine of those ten years, the conversation with a customer always ended the same way: "Looks great. Where does it actually fit in our business?"

The answer that question wanted, all along, was: "Pick a job that needs an owner. Give me the outcome. I will give you back a teammate." Not a tool. Not a model. Not a fleet. A worker, with a name, accountable for a piece of work.

We could not have started here in 2016. The technology could not hold a job. Even at the start of 2024, the agent layer could not consistently close the loop on a multi-step task against a customer’s real stack. So we built five things and shipped them. And we listened to the part the customer kept rephrasing.

What the customer kept rephrasing was always the same sentence, dressed differently each year: make this responsible for an outcome.

In 2026, for the first time, we can. The model can plan. The infrastructure can carry the plumbing. The memory can compound. The review layer can catch the 1% that matters. The category — finally — can support a unit of account that is not "tokens" or "bots," but the work.

So the bet is no longer about how much we can automate. It is about how clearly the business can define the outcome it wants an AI to own. The constraint is no longer what the AI can do. The constraint is what we can scope a hire around. That is a much better problem to have than the one we started with.

If you do not know where to start being AI-native, start with one employee. Not a strategy. Not a roadmap. Not a chief AI officer. One employee, one job, one outcome that lands on a board slide on Monday. The strategy follows the artifact. The fluency follows the work.

If you have been waiting for the version of AI that does not need you to translate it into your org chart — this is that version. We took ten years getting here. The first job is forty-eight hours.