Takayuki Kawazoe

Solo founder building Codens (AI dev suite, 5 specialized agents) with a public per-token rate card so customers can compute task cost upfront. Sentry → fix PR + E2E test gen + PRD automation.

Osaka, Japan Joined on Apr 28, 2026 https://www.codens.ai/

Takayuki Kawazoe

May 11

"How one empty message poisoned an entire AI consultation (and the three-layer fix)"

#ai #claude #python #debugging

7 min read

Takayuki Kawazoe

May 10

"Persisting your real Chrome login across Playwright restarts on macOS"

#playwright #chrome #macos #automation

7 min read

Takayuki Kawazoe

May 8

"Why your long-running AI agent feels broken (even when it isn't)"

#ai #agents #ux #python

11 min read

Takayuki Kawazoe

May 7

"Killing the 5-MCP setup tax with one PyPI package and Device Code Flow"

#mcp #claude #python #oauth

10 min read

Takayuki Kawazoe

May 6

How we ended up running one product with 2-3 people, after building our own dev harness

#ai #productivity #devops #saas

11 min read

Takayuki Kawazoe

May 6

Codens vs Devin vs Cursor Composer vs Sweep — picking the AI coding agent that matches your bottleneck

#ai #devops #claude #productivity

10 min read

Takayuki Kawazoe

May 6

We shipped a pricing page that was 50,000x off from the meter — a SaaS pricing post-mortem

#saas #pricing #stripe #ai

11 min read

Takayuki Kawazoe

May 5

The unit you pass between agents is the architecture — Purple to Blue with the implementation diff

#ai #claude #testing #playwright

12 min read

Takayuki Kawazoe

May 3

Claude Code Skills vs deterministic verify commands — same checks, very different ergonomics

#claude #ai #devops #testing

11 min read

Takayuki Kawazoe

May 2

I shipped a regex to catch AI agents committing only half the work. Then I ripped it back out.

#ai #claude #python #testing

14 min read

Takayuki Kawazoe

May 2

3 MCP server failure modes that bit us in production, and how we ship around them

#mcp #claude #python #devops

11 min read

Takayuki Kawazoe

May 1

One SSE connection, N upstream agents — the stream relay we shipped

#sse #redis #fastapi #python

12 min read

Takayuki Kawazoe

May 1

Why we built our own credit ledger instead of using Stripe metered billing

#stripe #billing #postgres #architecture

11 min read

Takayuki Kawazoe

Apr 30

I let my AI agents be advisory-only. Here's the rules-first PR risk engine I shipped instead.

#ai #claude #github #devops

7 min read

Takayuki Kawazoe

Apr 29

Stop AI from hallucinating E2E test selectors — code analysis + live browser exploration via Claude Agent SDK and 2 MCP servers

#ai #claude #mcp #playwright

6 min read

DEV Community

Takayuki Kawazoe

Badges

Writing Debut

"How one empty message poisoned an entire AI consultation (and the three-layer fix)"

"Persisting your real Chrome login across Playwright restarts on macOS"

"Why your long-running AI agent feels broken (even when it isn't)"

"Killing the 5-MCP setup tax with one PyPI package and Device Code Flow"

How we ended up running one product with 2-3 people, after building our own dev harness

Codens vs Devin vs Cursor Composer vs Sweep — picking the AI coding agent that matches your bottleneck

We shipped a pricing page that was 50,000x off from the meter — a SaaS pricing post-mortem

The unit you pass between agents is the architecture — Purple to Blue with the implementation diff

Claude Code Skills vs deterministic verify commands — same checks, very different ergonomics

I shipped a regex to catch AI agents committing only half the work. Then I ripped it back out.

3 MCP server failure modes that bit us in production, and how we ship around them

One SSE connection, N upstream agents — the stream relay we shipped

Why we built our own credit ledger instead of using Stripe metered billing

I let my AI agents be advisory-only. Here's the rules-first PR risk engine I shipped instead.

Stop AI from hallucinating E2E test selectors — code analysis + live browser exploration via Claude Agent SDK and 2 MCP servers