Staff Software Engineer. I help senior engineers think and design like architects. Writing & courses @ https://teapawt.net/t.co/R5AbMlF69E
London, England
Joined December 2010
- Tweets3.2K
- Following1K
- Followers372
- Likes5.3K
Gaz retweeted
ثبتنام هکاتان دیوان داره تموم میشه.
قراره کلی از همدیگه یاد بگیریم.
اپلای اینجا:
hackathon.divan.ac/
Ask a team what they are building and you hear "an agent".
Ask which steps it takes and most of them can draw it on a whiteboard.
That is a workflow. Cheaper, and you can forecast the bill.
gazar.dev/podcast/almost-not…
I use AI every day and I'm not going back.
What it quietly took is the part of me that writes code from an empty file.
So I built the small fix: 10 minutes, one TypeScript problem, hidden tests, no pasting, and it comes back on a schedule.
Free: coding.gazar.dev
You know your context window's size.
You have never counted what it costs per turn.
Three MCP servers. Twelve tools. ~3,000 tokens spent before the user types a word.
Your tool schemas are re-sent on every request.
youtu.be/PVjrOOyc0sE
Market leaders don't fail by doing the wrong thing. They fail by doing the right thing: listening to their best customers.
Meanwhile the cheaper, worse tool gets better fast and eats the market.
The Innovator's Dilemma, explained for engineers:
gazar.dev/reviews/innovators…
Your AI agent is holding your SSH key right now.
A tool with no dependencies, no install script, signed and pinned, asking for one permission: read files.
It still reads your .env and posts it out.
No code. Three sentences of English.
youtu.be/ACanodcYSig
Most slow code I see in review is not a slow function. It is the right function called too many times.
A loop, with a second loop hidden inside it. It hides because it is spelled find, includes, or filter.
12 of them, and the fix:
gazar.dev/algorithm/algorith…
Same job. Five fields out of a support email.
qwen2.5:3b on my laptop: $0
gpt-4o-mini: $57 per 1M calls
gpt-4o: $950
All three got it right.
The laptop was also 18x SLOWER, not faster.
"Local is faster" is a claim worth retiring.
youtu.be/VenyARBAOjc
Most TypeScript style arguments are really about where you are paying: compile time, bundle size, or runtime shape.
`as` on an API response pays at runtime, in production, far from the cast.
8 anti-patterns and what to do instead:
gazar.dev/typescript/typescr…
Ask a coding agent for "rate limiting on the API" and you get a confident diff back.
It also quietly decided the algorithm, which header identifies the caller, and what the reset value means.
Those were your decisions. You gave them away in one sentence.
The useful half of spec-kit is not the spec, it is clarify.
It interrogates the request before any code exists, so you answer the ambiguity in the repo, not in review.
Can a caller farm fresh quota by rotating X-Client-Id? Answer once. Every future task inherits it.
A spec without a fence is a suggestion.
The other half is one dependency-free file that reads the changed files out of git and refuses a diff that went out of bounds.
The whole walkthrough, on a real codebase:
gazar.dev/podcast/spec-drive…
How I Find Problems Worth Solving gazar.dev/leadership/how-i-f…
Rust has already taken the layer beneath your job. It has not taken the job.
Where the evidence points, and where I would still stay in TypeScript:
gazar.dev/system-design/is-i…
Microsoft wanted the TypeScript compiler 10x faster.
They picked Go, not Rust.
The constraint was not performance and it was not safety: the compiler's data structures are heavily cyclic, and cyclic graphs fight single ownership.
Architecture Patterns for AI Agents: The 5 Shapes That Actually Ship youtu.be/StOglU_9b7E?si=upl-… via @YouTube
Situational Awareness: The Decade Ahead to AGI by Leopold Aschenbrenner gazar.dev/reviews/situationa…
Why I Route Every AI Call Through OpenRouter Instead of Direct Endpoints gazar.dev/artificial-intelli…
System Design for AI Agents: Senior vs Staff!
Check out this free Lightning Lesson on Maven 👇
maven.com/p/eb3b31/system-de…
🎟️ 3 scholarship seats available
My next Maven cohort, Production-Ready Systems with LLMs & Agents, starts Tuesday.
👉 Apply by Monday, 11:59pm (London): gazar.dev/scholarship
