Angelo Applewhite (@builtbyangelo) header
Angelo Applewhite (@builtbyangelo) profile picture

Ex-SDR turned AI-native dev | Building software and products with coding agents | Agentic workflows

Charlotte, NC
Joined December 2020
Photo by Angelo Applewhite (@builtbyangelo)Photo by Angelo Applewhite (@builtbyangelo)Photo by Angelo Applewhite (@builtbyangelo)Photo by Angelo Applewhite (@builtbyangelo)Photo by Angelo Applewhite (@builtbyangelo)Photo by Angelo Applewhite (@builtbyangelo)Photo by Angelo Applewhite (@builtbyangelo)Photo by Angelo Applewhite (@builtbyangelo)Photo by Angelo Applewhite (@builtbyangelo)Photo by Angelo Applewhite (@builtbyangelo)
Becoming an AI-native developer is a path I didn't see myself on over a year ago. I've spent the last 16 months learning, building, and shipping products and applications. That meant long nights fixing codebase errors, deployment issues, sloppy executions, agentic hallucinations, and the classic "it works on my machine" problem. Agentic development is powerful when you leverage it correctly. It can give one focused engineer the output of a full development agency. You can instruct agents to run rigorous code reviews around the clock and even merge code without constant oversight. The tools move fast, but the fundamentals still matter. I've learned more by shipping and fixing real issues than by any tutorial. Here's what I've built so far: engineeredxangelo.dev
2
1
1
3,024
An agent reviewing its own work in the same session mostly grades the diff against its own reasoning, since its plan is still in context. Claude Code's docs say a fresh reviewer helps because it hasn't seen that reasoning. So I split the job between two agents. One is the builder, which writes and fixes the code. The other is the checker. It owns code review and hunts for bugs the builder missed. Findings go back to the builder. The loop ends when lint and tests pass and the checker's architecture review flags nothing.
1
23
In OpenAI's GPT-6.2 Astra and Sol post, past the benchmarks: in two tests both models were more likely not to mention what blocked their work, like a missing file. They added instructions in Codex to cut that down. If you run them in another harness, I'd add my own.
2
42
Easy to miss in Claude Code: an @ import in your instruction file isn't a link. The whole file gets loaded into context every session, needed or not. If you just want the agent to know a doc exists, put the path in backticks so it stays plain text.
21
Starting a series on rebuilding ContextOps with agents. Before any code I wrote a spec: architecture, data model, security, a phased build plan, and the questions still open. Agents make lots of small calls and I wanted them made against something written.
31
I get why people run a bunch of agents in parallel, I'm just not sold on it for a build I care about. Every agent works from its own slice of context and makes small calls the others never see. One well structured agent with a real plan keeps it all consistent with minimal debt for me.
38
Once in a while, write a small function yourself, then have the agent write its own and compare. Where they differ, ask it why. Often it knows something you didn't, and when it doesn't you get to push back on it, which is where a lot of the learning happens.
1
46
Agent summaries love "all tests pass." Sometimes the whole suite ran. Sometimes it ran the one file it touched and rounded up. Asking for the exact command and the last few lines of output clears that up fast. Anyone just having a Stop hook run the suite instead?
3
1
2
86
"If you still think that reading the code is the solution to all these problems, then you're going to have a whole separate set of problems." (Theo) Agree. Set the agent up to measure its own work and gate merges on checks, and that's the workflow I want.
36
Before you end a long agent session, ask it for a short note on what it tried and ruled out, and why. Most decisions show up in the code eventually, but the dead ends never do, so the next session tends to walk right back into one you already paid for.
2
1
57
When an agent uses an API or option you've never seen, ask it to show the docs or source it got that from. You learn the library, and if it can't point to anything, that's often where an invented option slipped in. Do you check, or let the type checker catch it?
27
Agent diffs that rename things and change behavior at once bury the behavior change. Ask for the refactor first as its own commit, without touching what the tests check, then the real change on top. If tests pass after the refactor, skim it and really review the second.
1
2
158
Coding Agents Tip: After an agent writes tests, have it break the code on purpose. Flip a comparison, delete a null check. One at a time, rerun the suite each time. If it stays green, those tests aren't covering what you think. Tools like Stryker automate this, but a few by hand tell you plenty.
4
1
4
217
Agents overbuild. Ask for a fix and you get the fix, plus a config option nobody wanted and a fallback for a case that can't really happen. Before review, try asking the agent which parts of the diff could go with the task still done. Cut those, then review what's left.
28
When an agent fixes a bug, ask why it happened before you read the diff. If it can walk from the input to the bad output, you can check the fix against that. If the explanation stays vague, it probably patched where the error showed up and never found where it started.
1
38
Chrome, Edge, Firefox and MDN just shipped 203 web guides for coding agents. They don't get dumped into context. The skill has the agent search a CLI and pull back only the guide for the task it's on. I'd copy that for any rules doc too long to load every session.
4
1
2
179
When an agent's PR touches package.json, I'd read that part before anything else. Every other line in the diff is code the agent wrote and you can review. A new dependency is code nobody on the team has read, plus whatever comes along with it in the lockfile.
26
Read the fine print on Anthropic's self-hosted Managed Agents. You run the server, sandboxes, credentials and session history, but model calls still go out. For a coding agent that history is mostly file contents and command output, the part I'd want on my own infra anyway.
1
29
When a coding agent flips the moment you push back, it isn't necessarily agreeing with you. Models fold under pushback even when they were right. Before it rewrites, ask what would have to be true for its original approach to be correct. Sometimes it's a constraint you forgot.
2
1
85
Google ran the same tasks with and without each skill, then deleted any skill that didn't beat the baseline. They rerun the check after a model upgrade, since a newer model may already know the workflow. I'd start doing this on my own skills, especially after a model switch.
29
Microsoft swapped the engine behind Copilot Chat in Visual Studio. Same models, new harness. On 157 real .NET tasks it resolved up to 21 more, and a task only counted if it didn't break tests that already passed. How much of what we call model quality is really the harness?
1
146