The mythical AI-agent month(thetechenabler.substack.com)
thetechenabler.substack.com
The mythical AI-agent month
https://thetechenabler.substack.com/p/the-mythical-ai-agent-month
6 comments
This exactly matches my experience. I also suspected that it was my higher threshold for code quality but Ai generated code is just not worth adding to a project without very strict reviews, unless it's non production and I want to fully give the project over to Ai
You can keep burning tokens until it complies. Thats what I do and I get good results. I do often have to spend a day just thinking through the prompt, but then again coding rarely was the bottleneck. But AI is very good at doing a refactor as well, tedious stuff like constructor juggling. Thing is code is to be written for humans first, no matter if the author is human or AI.
> You can keep burning tokens until it complies. Thats what I do and I get good results.
Like "it compiles" is literally the lowest bar, not "good results."
Like "it compiles" is literally the lowest bar, not "good results."
Depends on the language and the way code is written. Actually if you can run the code locally, getting things to compile is a bottleneck in Java especially for constructor juggling. Once code compiles, iterations do the rest. Iterations are better done by hand I think, we can get feedback on the code while keeping things fresh in our heads. For critical stuff where coding = thinking, its just faster to code.
Brooks’s law was about humans, but the spirit still applies: the real bottleneck is how much of the system a human can keep in their head at once. Most agent setups I’ve seen don’t reduce that load, they increase it: more surface area, more glue code, more boilerplate that nobody fully understands.
The only places they’ve worked well for me are roles where you can afford to throw away 80% of the output (migrations, test stubs, scaffolding) and keep a tight human-owned core. Treat agents like interns you fire every night, not teammates you trust with the architecture.
The only places they’ve worked well for me are roles where you can afford to throw away 80% of the output (migrations, test stubs, scaffolding) and keep a tight human-owned core. Treat agents like interns you fire every night, not teammates you trust with the architecture.
I agree that Brooks's Law applies here, but I think it bites at a different level than suggested.
An engineer coordinating AI agents can achieve coherent architecture. The bottleneck is less about human-AI coordination; it's that the inert organizational structures won't adapt.
The engineer now has to coordinate with AI agents and all the legacy coordinating roles that were designed for a different era. All these roles still demand their slice of attention, except now there are more coordination points, not fewer - AI agents themselves, new AI governance roles, AI policy committees, compliance officers, security assessments...
An engineer coordinating AI agents can achieve coherent architecture. The bottleneck is less about human-AI coordination; it's that the inert organizational structures won't adapt.
The engineer now has to coordinate with AI agents and all the legacy coordinating roles that were designed for a different era. All these roles still demand their slice of attention, except now there are more coordination points, not fewer - AI agents themselves, new AI governance roles, AI policy committees, compliance officers, security assessments...