The AI writes the code.
Keeping it maintainable is still your job.
Claude Code, Cursor and OpenCode all work beautifully in a demo and then drift on your actual codebase. The problem is rarely the tool — it is how the tool is wired into your particular repo, and that is the part a video cannot reach.
1:1 · 90 minutes · two sessions a week
Where AI coding actually goes wrong
Not in the demo. In week three, on the codebase you have to keep working in.
Context drift
The agent starts on-task and ends somewhere else. Your instructions and context files were written for a smaller repo and never updated as it grew.
Review costs more than writing
Generated code takes longer to read and verify than writing it yourself would have taken. The speed shows up in the first draft and disappears in review.
Quiet architecture drift
Nothing breaks. Abstractions get weaker, tests thin out, and the shape of the codebase moves a little every week until it is hard to change.
Tools keep moving
A new model, a new agent, a new config format. What survives the churn is a process, and most people never had one to begin with.
Free and open source
OpenAgents Control
An agent framework for plan-first development: the agent proposes a plan, you approve it, then it executes step by step with validation and review built in. Multi-language, model-agnostic, and free.
TypeScript · Python · Go · Rust · any model
Say no at the Approve stage and the plan gets rewritten, not your repo.
- Discover — Reads the patterns already in your repo
- Propose — The whole plan, written out
- Approve — Your call. Nothing written yet.
- Execute — One component at a time, checked each step
- Validate — Tests and a security review
- Ship — Nothing left to refactor
curl -fsSL https://raw.githubusercontent.com/darrenhinde/OpenAgentsControl/main/install.sh | bash -s developerNeeds the OpenCode CLI — the installer sets it up if you do not have it.
See it on GitHub →The one thing I sell
The Workflow Review — 90 minutes, 1:1, on your actual work
Bring a codebase, a project you are planning, or an agent setup that keeps going wrong. We work on it together on a screen-share — I change things while you watch. You leave with a written plan.
- Ninety minutes on a screen-share, working on your actual situation — not a talk about workflows in general.
- If you bring a repo, we work in it: context files, agent instructions and review loop rewritten live, by me, while you watch.
- A written action plan typed during the call, in a shared doc you keep.
£150
one session · 90 minutes · nothing recurring
Book a Workflow ReviewIf I can't help you, I'll say so in the first 15 minutes and refund you in full. No argument, no forms.
The system underneath it
The tools will keep changing. This does not. It is what I use on my own projects, and what we install on yours in a review.
- 01
Plan
Turn the idea into a spec with constraints and acceptance criteria, before any code is generated.
- 02
Prompt
Give the tool the repo knowledge, examples and boundaries it needs so the output is repeatable rather than lucky.
- 03
Build
Work in controlled steps small enough that you can still tell what changed and why.
- 04
Review
Catch weak abstractions, thin tests and security gaps before they are load-bearing.
- 05
Ship
Deploy only what held up under review.
From the channel
These are comments on the videos, not testimonials from paid sessions — the Workflow Review is new. Judge the teaching for yourself first.
I am glad I came across your channel. Your wisdom is much appreciated! I am learning a lot more than people just telling me what the newest AI tool is or showing off their latest work without really ever even telling you how to do anything.
There's so much uncertainty and AI hype out there that it's difficult to find someone I trust to explore, distill, and present things in a clear way. I look forward to more!
Great content! Just getting my feet wet with AI coding. I found this extremely helpful for understanding how things actually work. I'd love to see more on coding small projects with open source models.
Build notes
What I actually changed that week, what broke, and what I would do differently — the same material that ends up in OAC and in the sessions. No schedule, no sequence, no pitch.