SKF × studio91.ai studio91.ai
Scene 1 / 10
Chat
Produce
Remember
Act
Embed
Agentic
Govern
From Chatbot to Autonomous Agent

How did we get from a chatbot to an autonomous agent in about five years?

Seven rungs. Each one stacks on the last — nothing gets replaced.

Chat
2020–22
Produce
2023
Remember
2024
Act
2024–25
Embed
2024–25
Agentic
2025
Govern
Now
Stage 1 · Chat

Text in, text out — a conversation

The ChatGPT moment: AI as something you talk to.

Summarize this week's plant downtime report.
Here's a three-line summary: downtime rose 4% week-over-week, mainly driven by one conveyor fault at the Pune line…
One interface. Ask, and answer.
Stage 2 · Produce

Inputs expand. Outputs expand.

Click any icon to see an example.

In
Out
Click an icon above to see an example.
It stopped being a text writer and became a production team.
Stage 3 · Remember

It stopped being amnesiac — it now holds context

Click a card to see it in action.

Click a card above to see it in action.
It stopped being amnesiac — it now holds context.
Stage 4 · Act

It got hands

It can reach outside its own head — not just talk.

Web / Browser
Email
Database
CRM
Internal tools
AI
Connected via MCP (Model Context Protocol) — a common way for models to plug into outside systems.
It can fetch, read, and do — not just talk.
Stage 5 · Embed

It shows up inside the tools you already use

The opposite of Act: instead of reaching out, it's built in.

AI
Excel
AI
Notion
AI
Teams / Slack
AI
Slides
AI
Code editor
It stopped being a place you go to — and started being a feature already there.
Stage 6 · Run — Agentic

From assistant to operator

It runs independently for a stretch — minimal supervision.

Agent / Agentic: an AI system that pursues a goal on its own — planning, using tools, and checking its own work — rather than answering one prompt at a time.
1
Goal
2
Plan
3
Use tools
4
Check work
Takes a goal, plans the steps, uses tools, checks its own work — and keeps going.
Stage 7 · Govern

Now it needs guardrails

Because agents act on their own, you define the boundaries — and score the result before you trust it.

Success testsDid it solve the task correctly?
Boundary conditionsWhat's it allowed to touch, spend, or send?
Stop conditionsWhen does it halt and hand back to a human?
Most enterprise AI pilots don't fail on model quality — they fail on the lack of this layer.
The Arc

Where does your function sit today?

Chat
Produce
Remember
Act
Embed
Agentic
Govern

And what's the next rung?

Quick Check

Let's see what landed

Presenter note