What I'm doing

Into a fully AI-coded future

AI has fundamentally changed how software engineers build software. The near-term plan is to explore the following areas.

01

Preparing

Tests for hot AI coding claims

Run it against a real task

AI coding gets a new hot take every few months. People started promoting general workflow skills early in the year; by July, a wave of claims said those skills should be abandoned. Rather than decide who sounds more convincing, the plan is to run the tasks.

Planned test

The test will use short-, medium-, and long-horizon tasks, comparing current flagship and Flash models with equivalent models from when a skill became popular. The same task will run in an opinionated harness such as Codex and a minimal one such as pi. The comparison should show when the claim holds, how many extra tokens it costs, whether success rates move, and how much of the result comes from the harness. These tasks and metrics should eventually become a reusable benchmarking tool across models, skills, and harnesses.

  • Short / medium / long tasks
  • Flagship / Flash models
  • Codex / pi
  • Benchmarking tool

02

Ongoing

Software engineering for agent-written code

What should Review look like now?

AI coding increasingly looks like a new programming language. If agents write most of the code, line-by-line human review should not remain the default. That does not mean removing Review; it means finding a better way to do it.

Direction to test

Move Review a few levels up: architecture, data flow, state boundaries, and runtime results. Then test TDD, SDD, and whatever other practices people are using against real tasks. This area ultimately needs a new Review methodology for AI coding and the tools to support it.

  • Architecture
  • Data flow
  • TDD / SDD
  • Review methods / tools

03

Prototyping

Small agents people can use

Start with toys that actually run

The third kind is small Web Agents that someone can use immediately. The first target is the “working toy” stage instead of a general platform.

What comes first

One example is an AI version of AdGuard whose rules are generated and maintained by AI. Another is the “笨蛋谢琪 Agent” on this site: visitors ask about the projects, code, and experience documented here, and it generates a UI that makes the answer easier to understand. The first two areas test whether the methods work; the third proves I can build Agent-native products.

  • 笨蛋谢琪 Agent
  • AI AdGuard
  • Immediately useful
  • Get it running