Skip to main content
ariefrse.my
HomeSkillsProjectsAI WorkWritingContact
Resume

Writing

Writing

Notes and articles from real production work. The incidents, the fixes, and what each one taught me.

How it connects

Most of these pieces reference each other, because most of them are the same lesson arriving from a different direction. Hover a node to see what it links to.

Agentic engineeringPracticeInfrastructureNext.js25 pieces · 49 links · drag to rearrange

Notes

Short write-ups of things that broke in production and what it took to find them.

Shipline left confidence scores out. Jev is why they are coming back

Sep 30, 2026 · 7 min read

Note · Agentic engineering

A confidence score written in prose is decoration. A typed verdict with a calibrated number can stop a pipeline. The difference is who consumes it.

I built my agent workflow into a tool so it stops being optional

Sep 14, 2026 · 6 min read

Note · Agentic engineering

Knowing the right process and following it under deadline are different skills. One of them can be automated.

The redesign that landed on one page out of eight

Sep 10, 2026 · 4 min read

Note · Practice

If a design decision lives in more than one file, it is not a decision yet. It is a coincidence waiting to drift.

The sweep that found managers could pay other teams' claims

Jul 22, 2026 · 5 min read

Note · Agentic engineering

The scariest findings were not crashes or wrong numbers. They were correct-looking actions performed by the wrong person.

An HR platform where seven in ten commits came from an agent

Jul 22, 2026 · 6 min read

Note · Agentic engineering

Seven in ten commits came from an agent. Zero in ten decisions about what "correct" means for the business did.

A green build is not a deploy: making CI check the live site

Jul 16, 2026 · 4 min read

Note · Infrastructure

Every stage of my pipeline vouched for itself and nobody vouched for the result. The first real run of the fix caught ten days of deploys that had gone nowhere.

The settings that did nothing and the icons that didn't exist

Jul 16, 2026 · 5 min read

Note · Next.js

Configuration is a list of claims, and nothing checks claims by default. Two failures with the same shape: declared but never delivered.

Everything fails silently

Jul 16, 2026 · 6 min read

Note · Practice

Green is not evidence. It is the absence of evidence. A running gallery of production failures that passed every check that existed.

Hosting a company's HR system on a PC that never turns off

Jul 7, 2026 · 8 min read

Note · Infrastructure

A production pilot on one desktop machine, zero hosting bill, real TLS and a hidden origin. What that buys, and what it honestly costs.

Why I deleted my most impressive-looking case study

Jul 6, 2026 · 3 min read

Note · Practice

One unbacked claim is not one bad item in a list. It is a discount applied to the whole list.

The menu link that did nothing when clicked

Jul 6, 2026 · 5 min read

Note · Next.js

A shared "render everything through Link" abstraction is a trap the moment one item is off-origin. The click is intercepted and then nothing happens at all.

How my website quietly lost its security padlock

Jul 6, 2026 · 4 min read

Note · Infrastructure

Every trailing-slash redirect was downgrading visitors from HTTPS to plain HTTP. Browsers follow that without a word, so nothing ever looked broken.

How my site updates itself minutes after I save my work

Jul 6, 2026 · 5 min read

Note · Infrastructure

A push-to-live loop with no SSH step, and the two footguns that each cost me a debugging session. Both failed invisibly from the outside.

Letting an AI hunt for the bugs humans skim past

Jul 3, 2026 · 5 min read

Note · Agentic engineering

Parity bugs are exactly what humans skim and agents grind through. The most valuable finding was the fix a human had to revert.

Teaching AI assistants to remember past mistakes

Jul 3, 2026 · 4 min read

Note · Agentic engineering

Agents forget between sessions. The repo does not. Every regression a reviewer catches becomes a constraint the next session starts with.

Upgrading a website's engine: the bugs that don't crash

Jul 3, 2026 · 5 min read

Note · Next.js

Every project page on my portfolio rendered a 404 for months while returning HTTP 200. On this bug class, "it builds and the page loads" is what failure looks like.

Running agents in parallel is about waiting, not typing

Jul 3, 2026 · 4 min read

Note · Agentic engineering

The productivity is not in agents typing faster. It is in never sitting idle while something runs.

Write the plan before the code

Jul 3, 2026 · 4 min read

Note · Agentic engineering

One day of writing notes bought a 196-file MVP the next day, because the agent spent zero time guessing what I meant.

Articles

Longer pieces on engineering practice and agent-assisted delivery.

The prompt I left for whoever takes over

Sep 29, 2026 · 6 min read

Article · Practice

I handed over systems I had run alone. The dangerous first week is not the one with no documentation, it is the one where someone acts confidently on a stale mental model. Here is the prompt I left, with the specifics stripped out so you can use it.

System One models: the case for an AI that never writes a sentence

Sep 20, 2026 · 7 min read

Article · Agentic engineering

Typesafe shipped a frontier model that returns a typed value and a calibrated confidence score instead of text. No parsing, no malformed JSON, no hallucinated enum member.

Money-safe webhooks: idempotency, exact decimals, and a balance invariant

Sep 10, 2026 · 7 min read

Article · Practice

The bugs that hurt in invoicing are never crashes. They are the ones where the system is confidently wrong.

How AI changes the software development lifecycle

Jul 22, 2026 · 8 min read

Article · Agentic engineering

The lifecycle has not got shorter. It has got lopsided: cheap in the middle, expensive at both edges.

A beginner’s guide to AI-powered development

Feb 11, 2026 · 10 min read

Article · Agentic engineering

AI did not replace developers. It replaced the boring parts. The rule that keeps you safe: if you cannot explain the code, do not ship it.

Chain of Thought meets Agent Teams in Claude Code

Feb 11, 2026 · 7 min read

Article · Agentic engineering

Chain of Thought is the plan and Agent Teams is the execution. They are two layers of one system, not two techniques.

MCP: the protocol every agentic engineer needs to know

Feb 11, 2026 · 8 min read

Article · Agentic engineering

Connecting one agent to N tools used to mean N one-off integrations. MCP is the interoperability layer that removes the multiplication.

© 2026 Muhammad Arief Roseli

GitHubLinkedInSnapInvois