From chatbots to agents
Most people's experience of AI is a chatbot: you ask, it answers. The next step is already arriving — agentic AI that takes a goal and does the work across several steps, using tools and files, while you supervise. Here's what's changing, why it matters, and what stays the same.
The chatbot: a brilliant assistant in a text box.
You type a question or a request; it replies in text. General-purpose assistants like ChatGPT, Claude, Mistral and Gemini are excellent at drafting, summarising, explaining and answering. But they wait for you, work one turn at a time, and only know what you paste in. They help with each step — you still do the steps.
The agent: AI that acts, not just answers.
An agent takes a goal, makes a plan, and carries it out over many steps — using tools, reading and writing files, calling other software, and checking its own work — coming back to you for direction and sign-off rather than for every keystroke. The clearest example so far is in software: tools like Claude Code let someone describe what they want and have the AI carry out the steps to deliver it. The same pattern — set a goal, supervise the work — is spreading into everyday knowledge work.
You do the steps
You break the job down, prompt for each piece, copy results between tools, and stitch it together. The AI assists; you drive.
It does the steps
You set the goal and the boundaries; it plans, acts across tools, and returns a result for you to check. You supervise; it drives.
The real change is what you hand over. It moves from "a question" to "a task" — from asking for a draft and doing the rest yourself, to delegating the whole job and reviewing the result.
More leverage — and more to get right.
Two things grow at once. The upside: far more gets done with less manual effort. The responsibility: an AI that can act needs clear boundaries — what it may touch, where a human must approve, and how you check what it did. That's why oversight and governance matter more as AI grows more capable, not less. The good news: you don't have to predict the tools to be ready for them.
The tools will keep changing. These skills won't.
Whether you're prompting a chatbot or supervising an agent, four competencies decide whether AI helps or hurts. They're Kramer Consulting's AI-fluency framework — our adaptation of the widely used four-dimension model (Anthropic; Dakan & Feller):
Decide what to hand over
Work out what should go to AI and what stays human — the judgement calls, the relationships, the final say.
Set the goal well
Give clear instructions: the objective, the context, the constraints, and what a good result looks like. Vague in, vague out.
Check the work
Read the output critically — facts, logic, tone, fit. The more an AI can do unsupervised, the more this matters.
Take responsibility
You are accountable for AI-assisted work. "The AI did it" is never an answer to a client, a regulator or a colleague.
Two formulas make Direction concrete: for a chatbot, R-T-F-C — Role, Task, Format, Context. For an agent, G-C-O-V — Goal, Context, Order, Validate. The move from one to the other is the move from phrasing a request to scoping a job.
Get fluent with the assistants — then step up to agents.
There's a natural learning curve, and the Digital Learning Hub programme runs exactly that arc:
The Claude ecosystem
Get genuinely fluent with the assistants — including a no-code path into Claude Code. From introduction to in-depth practice.
Mistral & data control
Capable open models you can run where your data stays under your control — from introduction to practical workflows.
AI agents
Understand what agents are and why they matter — then build your first one in a hands-on day.
Already use ChatGPT or Microsoft Copilot? Start by getting them to know you — short, tailor-made ChatGPT and Copilot intensives cover the settings, projects and agents most people never set up.
Get ahead of the shift.
Open-enrolment seminars on Claude, Mistral and AI agents — small groups, hands-on from the first hour.
Related guides
Agentic AI design patterns
Most AI solutions you meet can be understood through seven recurring shapes — five workflows you design and two agents that work more on their own. Not an exhaustive taxonomy but a working selection, few enough to hold in your head. The one question that tells you which to reach for, a diagram and everyday examples for each, and the line from an agent you delegate to, to why it needs governance.
Read the guide Directing agentsDelegate the goal, not the task
A chatbot needs a fresh prompt every time; an agent needs a goal and the room to reach it. The G-C-O-V method for briefing agents — and the real unlock: building the brief into your setup so you stop re-typing it.
Read the guide GovernanceAI governance, explained
What it is, why it pays for itself, and how to make it live in your people instead of a binder nobody reads.
Read the guide