Written here, published there.
Everything Argot writes goes out under the client's name, on the client's blog. This is the work, one masthead at a time.
Building a C compiler with a team of parallel Claudes
Editing, for Anthropic's engineering blog
A researcher's account of running a team of Claude agents in parallel to build a C compiler, and what the experiment showed about long-running autonomous software work. Argot edited the draft: kept the voice, cut the detours, checked every figure against the run logs.
Read it on anthropic.comA guide to Terraform secrets management
How-to guide, for Infisical
Why every secret Terraform touches ends up in plaintext state, and how ephemeral resources, OIDC in CI, dynamic secrets and rotation keep real values out of state and Git. Leaks a secret into state on purpose in the first code block, so the reader sees the problem before the fix.
Read it on infisical.comGetting started with Claude in Excel
Video script, for Claude's YouTube channel
A getting-started video for Claude in Excel: reading a whole workbook, explaining formulas with cell-level citations, and changing assumptions without breaking the model. Written to be spoken: short lines, one idea per shot, nothing that needs a second read.
Watch it on youtube.comHow to deploy Whisper to transcribe audio in seconds
Tutorial, for Modal
Transcribes a podcast with OpenAI's Whisper on an H100 in a few lines of Python, after a plain account of how the model works and the ways to host it. Every command was run on Modal before it was written down.
Read it on modal.comBillion-line days demand billion-test nights
Opinion, for Momentic
Developers now accept about a billion lines of AI-generated code a day. The essay argues that testing has to grow at the same rate, generated from the same specifications as the code. An argument under the CEO's name, built from a public number rather than a product claim.
Read it on momentic.aiHow to build an AI app to write better outreach emails
Tutorial, for Retool
Builds a Retool app that drafts personalized outreach emails, from the form interface to the AI action and prompt that write each draft. Argot built the app before writing a word, so every screenshot is from a working build.
Read it on retool.comHow to scale event-driven systems without compromising app stability
Explainer, for Stream
Why event-driven workloads fail differently under load, and how to prevent cascading failures, absorb traffic spikes and stay reliable across regions. No product until the failure modes are clear.
Read it on getstream.ioCommonJS is hurting JavaScript
Opinion, for Deno
Argues that CommonJS is holding the JavaScript ecosystem back and makes the case for moving to ES modules. An opinion under a company's name has to survive the comments, so it was written to be argued with.
Read it on deno.comBringing AI to sensor data: Newton on-prem and real-time
Explainer, for Archetype AI
How Newton, Archetype AI's foundation model for sensor data, runs on-premises and in real time for industrial sites that can't send their data to the cloud. The constraint comes first, data that can't leave the site, and the product only once the reader feels it.
Read it on archetypeai.ioSecure authentication in Next.js with email magic links
How-to guide, for Clerk
Explains how email magic links work, builds them from scratch in Next.js, then shows the same flow with Clerk. The from-scratch version comes first so the reader sees exactly what the product saves them.
Read it on clerk.comImage classification benchmark: Google, AWS, Hugging Face and Nyckel
Benchmark, for Nyckel
Four image classification services compared on accuracy, training time, latency, throughput and developer experience, across more than 100 trained models and seven datasets. Bylined to Argot. Over 100 models trained by a writer with no ML background, which was the point.
Read it on nyckel.com