wrhn

Recent submissions

Comment · 18 days ago
The code generated by the coding agents should not be reviewed, or tested by people. A human's working memory (the biological equivalent of an AI context window) holds about 4 to 7 items (or "chunks" of information) at one time. With 10s of PRs waiting in the pipeline to be reviewed, judged and tested, we're not only putting our SDLC systems under pressure, also ourselves and peers under pressure. CI/CD, reviews, testing, even git started to become bottlenecks and cooked under pressure. Even though you tackle one, there will be the next cycle which will become bottleneck under the pressure of the outcome. It's like hiring 100 engineers at once, of whom start to generate code rapidly from the first minute they're onboarded. Many systems are not ready for this scale.

This is not sustainable for the industry. People are already burning out, quitting the profession, or chasing an exit.

Agentic software development lifecycle should be covered completely by the agents for day to day work, where people build and maintain the specs, guardrails, and agentic engineering systems around the agents.

I've built https://letsship.ai with this vision, and it's the only sustainable way of working for me.

Comment · 2 months ago
Yes, curious. Will do!
Comment · 2 months ago
I understand where you're coming from. The research focuses on Builder configuration being variable and whether it affects metrics like cycle time and cost, which it shows the diff in detail.

I'm interested to see the total inference cost of a ticket type, that's why total cost is compared. Even though most of the cost is there in other stages, there's still huge difference even with one variable configuration in the loop.

The cost of each builder is also shown in mission breakdown, but good feedback for extending some of the tables, thanks!

Comment · 3 months ago
It's an orchestration layer on top of coding agents like Cursor and Claude Code. It uses coding agents with the same configuration that builders use.

SHIP users configure each specialist agent with a pointed harness and a model of choice, along with skills, memory and any tool integration. These agents then are operated to work together and deliver a reviewed, gated, tested outcome.

Comment · 5 months ago
I've built this course specifically for engineers, but I doubt if this is the right audience for it.

If you're an engineer, tell me why you're NOT interested in such a course/format please.

Live from the official Hacker News API

A modern reader for Hacker News · SvelteKit + Skeleton