AI agent development
Agents that use tools to complete multi-step tasks, with narrow permissions, checks after every step and a full record of what they did.
What an agent is, without the hype
An agent is a loop: a language model decides on the next action, calls a tool (search, read a file, run a query, call an API), looks at the result and decides again until the task is done or it gives up. That is useful when a task needs several steps that depend on each other. It is also where things go wrong in compounding ways, because every step can fail and the model will often continue confidently.
What gets built
- Narrow tools: each tool does one thing, validates its inputs, and has the least access it needs. Destructive actions require confirmation or are not available at all.
- Verification: after each meaningful step the agent’s work is checked by something other than the agent: tests, schema validation, a rule, or a person.
- Budgets and stopping rules: maximum steps, time and cost per task, and a clear “I could not do this” outcome instead of an endless loop.
- Traces: every decision, tool call and result recorded, so a failure can be diagnosed and the case added to the test set.
- Evaluation: a set of representative tasks run on every change, measuring success rate, cost and time, not just whether a demo worked once.
Good fits and poor fits
Good fits are tasks with clear success criteria and cheap verification: triaging and enriching tickets, multi-source research with citations, code changes checked by a test suite, data cleanup with validation rules.
Poor fits are tasks where nobody can tell whether the result is right, actions that are irreversible and expensive, and processes that are really fixed workflows. A fixed workflow is better built as ordinary software, with a model used only at the steps that need judgement.
How an engagement starts
By writing down the task, the tools it would need, and how success is checked. If success cannot be checked, that is the first problem to solve.
Have a project like this?
- I read your message and reply personally, normally within two working days.
- Any questions are clarified by email. A short call only if required.
- A written proposal for a fixed-scope first step, with a fixed price.
How an engagement works
Every project starts small and fixed in scope, so you see evidence before you commit to more.
Feasibility assessment
A short, fixed-scope look at your problem, your data and the options.
A written assessment and a clear recommendation, including "don't build this" when that is the honest answer.
Prototype
A working system on your real data, not a demo on someone else's.
The prototype plus an evaluation report with measured results.
Build
The production system, integrated with what you already run.
Tested, documented code that you own, with a proper handover.
Support
Monitoring, evaluation reruns and improvements as your data changes.
Agreed scope and response times, no lock-in.