---
title: "The Builder"
slug: sep-2026
date: 2026-09-16
description: "Agents importing Figma designs, writing to git, and running unsupervised for months. Plus TypeSafe's hallucination-proof model and a notch for coding tools."
edition: 7
canonical: https://lukestahl.io/newsletter/editions/sep-2026/
---

Agents, agents, agents. This month's issue is mostly agents doing things without being watched: agents importing Figma designs, agents writing to git without a human checking the diff, agents running for months without a check-in. Somewhere in there, TypeSafe shipped a model that can't hallucinate because it never generates strings in the first place. Make of that what you will.

Also inside: a macOS notch that tracks whether your six different coding assistants are working or stuck, and a dictionary for the fifty new terms you now have to know to talk about any of this. Dig in below.

## Blogs & Projects

- <img src="/blog/images/newsletters/figma-to-webflow-mcp.jpg" alt="Import a Figma design into Webflow with one prompt" class="li-thumb"><span class="li-text">**[Import a Figma design into Webflow with one prompt](https://webflow.com/blog/import-figma-to-webflow-mcp)**<br>One loose prompt with the Figma and Webflow MCP connectors turned an empty site into a full build — design tokens, classes, components, and assets all in place, no custom import tool required.<br>_Stefan Judis_</span>
- <img src="/assets/images/ai-model-guide.png" alt="AI Models Guide 2026" class="li-thumb"><span class="li-text">**[AI Models Guide 2026](https://lukestahl.io/ai-models-guide/)**<br>Not new, just a resource I keep updated weekly: every major model compared side by side, cost, context window, and what each one is actually best at.<br>_Luke Stahl_</span>

## Links I like

- **Git at any scale**: Cursor replaced quorum-based replication with a write-ahead log in S3, so pushes stay linearizable and reads scale to any number of replicas. [Read more](https://cursor.com/blog/git-at-any-scale)
  _Vicent Marti_
- **Build it live - an agentic Webflow workflow, start to finish**: Stefan Judis walks a live build with the Webflow MCP server at Webflow Conf, from setup to code components and CMS data. [Read more](https://webflow.com/webflowconf/session/build-it-live---an-agentic-webflow-workflow-start-to-finish)
- **Understand-Anything**: A Claude Code plugin that turns your codebase into a clickable knowledge graph, with guided tours and semantic search built in. [Read more](https://github.com/Egonex-AI/Understand-Anything)
  _EgonexAI_
- **ASD-STE100 skill**: A Claude Code skill that rewrites tool descriptions and error messages using aerospace maintenance-manual writing rules, so agents parsing the output can't misread it. [Read more](https://github.com/danyuchn/asd-ste100-skill)
  _Dustin Yuchen Teng_
- **Introducing Hybrid Compute on Mac**: The article provides insight into the integration of hybrid compute capabilities on Mac, detailing how it enhances performance by allowing seamless task distribution between local and cloud resources. [Read more](https://www.perplexity.ai/hub/blog/introducing-hybrid-compute-on-mac)
- **AI Evals: A Hands-On Guide for Product Teams**: Unit tests don't work on LLMs since output varies run to run. Teresa Torres breaks down error analysis and the four eval types you'd pick between instead. [Read more](https://www.producttalk.org/ai-evals/)
  _Teresa Torres_
- **Nango**: Handles auth, syncs, and MCP tool exposure for 1,000+ APIs, so agents and products connect to third-party services without you building the integration layer from scratch. [Read more](https://nango.dev/)
- **Introducing Cursor Projects**: A coordinator agent that delegates to subagents, holds context across months of work, and runs recurring tasks like migrations and design-system cleanup without being prompted. [Read more](https://cursor.com/blog/projects)
  _Alexi Robbins & Fredrika Lindh_
- **Codenotch**: A macOS notch showing usage limits across Claude Code, Cursor, Codex, and a dozen other coding assistants at once, with live status when a session finishes or gets stuck waiting on you. [Read more](https://github.com/vinzdg/codenotch/tree/main)
  _Vinz_
- **OpenAI's Agents API**: Codex's harness as a managed API, with long-session context management, tool search, and subagent support for parallel work. [Read more](https://openai.com/index/introducing-the-agents-api/)
- **AI Coding Dictionary**: 69 terms for agentic coding, from tokens and context windows to handoff patterns and failure modes like attention degradation and the "smart zone to dumb zone" drift. 3D graph at [aicodingdictionary.com](https://www.aicodingdictionary.com/). [Read more](https://www.aihero.dev/ai-coding-dictionary)
  _Matt Pocock_
- **Introducing System One Models & Jev**: Instead of generating strings token by token, Jev outputs typed structured decisions in parallel, no chat, no hallucination risk, at 70-500ms per call versus seconds for a comparable LLM. [Read more](https://typesafe.ai/blog/introducing-system-one-models-and-jev)
  _Diogo Almeida_
- **Bun 1.4**: +1,517 more Node.js tests pass, idle CPU drops 5x, memory drops up to 35%, and startup is 50% faster on Linux. Bun also rewrote its runtime from Zig to Rust. [Read more](https://bun.com/blog/bun-v1.4)
  _Jarred, Ciro, Dylan, Alistair & Sosuke_

## Technology was a mistake

- [What Happens If OpenAI Dies?](https://www.wheresyoured.at/what-happens-if-openai-dies/)
  _Ed Zitron_

## Technical & Developer Marketing Jobs

- [Senior Product Marketing Manager at Sourcegraph](https://job-boards.greenhouse.io/sourcegraph91/jobs/6132206004?gh_src=0cd0575f4us)
- [Developer Relations Social Media Lead at Render](https://jobs.ashbyhq.com/render/4611bde4-47ac-45fc-ab56-235489e52682?departmentId=ae208503-0f32-4d59-8c52-9622134d9fe3)
- [Director, Developer Marketing at Redis](https://redis.io/company/careers/director-developer-marketing-66bc9c95-1248-45f6-b3f1-d5c2d3994198/?ashby_jid=66bc9c95-1248-45f6-b3f1-d5c2d3994198)
- [Integrated Marketing Manager, Developers at OpenAI](https://openai.com/careers/integrated-marketing-manager-developers-san-francisco/)
- [Technical Product Marketer, Developer Platform at Workday](https://workday.wd5.myworkdayjobs.com/en-US/Workday/job/Technical-Product-Marketer--Developer-Platform_JR-0109407?source=Careers_Website)
- [Product Marketing Manager, Chrome for Developers at Google](https://www.google.com/about/careers/applications/jobs/results/110745968541868742-product-marketing-manager/)
- [Principal Product Marketing Manager, API at Synthesia](https://jobs.ashbyhq.com/synthesia/9126e40f-ff2c-4cf7-b696-67c3787983cc/application)
- [Head of Developer Technical Content, Cloud Product Marketing at Google](https://www.google.com/about/careers/applications/jobs/results/141745915055481542-head-of-developer-technical-content/)
- [Head of Startup & Developer Marketing at Rippling](https://ats.rippling.com/rippling/jobs/846ec886-676f-4b7c-a7c6-4b95558ae8af)
- [Senior Product Marketing Manager, Solutions at Deepgram](https://jobs.ashbyhq.com/Deepgram/0f597b67-3cd4-4d56-b951-d555031ee5cc)
- [Developer Marketing Manager at Meta](https://www.metacareers.com/profile/job_details/4372804292960620/)

## Folks to follow

- [Diogo Almeida](https://x.com/CompleteSkeptic)
- [Cassie Evans](https://x.com/cassiecodes)

---

That's it for September. If you're rethinking how your team talks to developers, I'm probably thinking about the same stuff. Find me on [LinkedIn](https://www.linkedin.com/in/lukestahl/) or [X](https://x.com/LukeStahl11).

— Luke