Productionai

AI agents in financial services need controls before autonomy

Financial-services AI agents can be useful, but autonomy without permissions, audit trails, segregation, evals, and rollback is just operational risk with a nicer interface.

Diagram showing green tests as one signal beside a Claude Code review packet with scope, command log, tool log, rollback note, and human approval

Claude Code green tests are not a review packet

A Claude Code run can make tests pass and still leave a reviewer with no usable evidence. Treat green tests as one signal, then require scope, command logs, tool use, assumptions, and rollback notes before merging agent work.

Diagram showing a Claude Code run stopping after repeated failures and producing a review packet instead of looping blindly

Claude Code needs a stop rule before more autonomy

Claude Code gets risky when a failed run keeps retrying without a stop rule. Use failure budgets, review packets, evals, and rollback notes before giving agents more autonomy.

Diagram showing a Claude Code team adoption runbook with task contract, scoped permissions, review packet, evals, and rollback

Claude Code team adoption needs a seatbelt runbook

Claude Code gets risky when teams roll it out through enthusiasm instead of a runbook. Start with task contracts, scoped permissions, review packets, evals, and rollback before widening autonomy.

Diagram showing how MCP tools widen Claude Code blast radius from local repository access to external systems unless boundaries, approval, logs, and rollback are added

Claude Code MCP tools need a blast radius, not a vibe check

MCP makes Claude Code more useful, but every server also widens the blast radius. Treat MCP tools as production access paths with allowlists, approval gates, call logs, and rollback notes.

Diagram showing the boundary where AI code generation moves from suggestion mode to action mode with files, commands, tools, and services

AI code generation gets risky when it becomes action

AI code generation is manageable when it suggests code. The risk changes when agents can edit files, run commands, call tools, and open pull requests.

Cover of Claude Code: Building Production Agents That Actually Scale by Thomas De Vos

Claude Code book is live on Amazon Kindle

Claude Code: Building Production Agents That Actually Scale is now live on Amazon Kindle. Here is who it is for and why I wrote it.

Claude Code rollback envelope diagram showing scope, agent change, and rollback path

Claude Code Rollback Plans Belong in the Prompt

If a Claude Code agent can change production-shaped code, the prompt should say how to undo the work. Rollback is not paperwork after the diff. It is part of the task boundary.

Claude Code agent cost loop diagram showing vague tasks, broad tools, repeated exploration, and no stop rule

Claude Code Agent Cost Loops Start as Workflow Bugs

Claude Code cost problems usually start before the model call: vague tasks, wide-open tools, repeated repo exploration, and no stop rule. Treat spend as a workflow bug, not just a pricing problem.

Claude Code evaluation loop showing capture, reduce, test, and change steps for failed agent runs

Claude Code Evals Should Start With Bad Runs

Production Claude Code evals should not begin with abstract benchmarks. Start with the agent runs that scared you, reduce them into replayable cases, and use them to tune permissions, prompts, tools, and review gates.