reference

features

1. promptingwhat predicts a frontier-model pick across 337 coding sessions, why the second reading disagrees with the first, and the model router updated in step

dossiers

sourced histories of the companies, labs, and products worth studying.

1. roam researcha sourced history of roam research and its founder conor white-sullivan, from the hype and the backlash to the unfinished idea of networked thought.
2. obsidiana sourced history of obsidian, its people, the discourse around it, and the file-over-app idea it is built on.
3. gumroada sourced history of gumroad and sahil lavingia's umbrella company, the discourse around them, and the small-company idea behind them.
4. everya sourced history of every, the media house and app suite, the discourse around it, and the bundle idea behind it.
5. ink & switcha sourced history of ink & switch, the independent lab behind automerge and the local-first software movement, and the tools-for-thought scene it seeded.
6. tldrawa sourced history of tldraw, steve ruiz's infinite-canvas company: the sdk business, the make-real demos, and the canvas-as-infrastructure thesis.
7. antithesisa sourced history of antithesis, the deterministic simulation-testing company from the foundationdb lineage, and its autonomous-testing idea.
8. midjourneya sourced history of midjourney, david holz's self-funded independent research lab, its community, and its models.
9. zeda sourced history of zed industries, the editor company founded by atom and tree-sitter veterans, from gpui to its idea of a fast collaborative editor.
10. core automationa sourced history of core automation, inc., the san francisco ai lab founded by former openai vp of research jerry tworek, not the industrial-automation firms.
11. morpha sourced history of morph, the apply-model company behind fast apply, from its lingtual pivot years to its subagent suite.
12. moving castlesa sourced history of moving castles, the berlin studio, its worlds, and the zero release.
13. typesafea sourced dossier on typesafe ai, the system-one lab behind jev: its founders, its launch, and the calibration question it leaves open.
14. cognitiona sourced history of cognition, scott wu's company behind the devin software-engineering agent, and its acquisition of windsurf.
15. deepseeka sourced history of deepseek, the open-weights ai lab, its founder liang wenfeng, and high-flyer, the hedge fund behind it.
16. andon labsa sourced history of andon labs, the ai-safety evaluation shop, covering project vend and its model-behavior experiments.
17. convergent researcha sourced history of convergent research, the nonprofit that incubates focused research organizations, and its cofounders adam marblestone and anastasia gamick.

lessons

technical columns written from the systems hraness runs.

1. agent infrastructuredurable sessions, orchestration, and review for coding agents.5 lessons · start with one writer per session: leases and fencing for durable agent sessions
2. local-first softwareowning state on the device, syncing on the owner's terms.4 lessons · start with local-first doesn't need crdts: a state model for single-user apps
3. knowledge systemsmarkdown vaults, retrieval, and evidence beside the code.4 lessons · start with markdown and git are the database; everything else is a derived index
4. web deliverymetadata, sitemaps, social images, and proof of which commit production serves.5 lessons · start with accept: text/markdown in production
5. design systemsprimitives, tokens, and brand catalogs shared across products.4 lessons · start with token cascades across a product family: one brand, many surfaces
6. browser mediaaudio, midi, and generative media in the browser.4 lessons · start with canvas as a control surface an agent can operate
7. accounts and commerceshared identity, billing, and marketplace payouts across products.4 lessons · start with one identity, many products: shared identity, per-app entitlement
8. the factoryone engineer, many products: monorepo operations and agent workflow.4 lessons · start with merge queues for agent-written commits: affected checks and serialized integration
9. rust systemsone cargo workspace, many small crates: ports, deterministic cores, and effectful shells.5 lessons · start with how we split a p2p system into small crates (and the rules that survived)
10. programming languagessmall languages built to be run, measured, and replayed.5 lessons · start with a policy language that decides what simulated organisms do
11. peer-to-peer systemsidentity, discovery, transport, and consensus without a server in the middle.5 lessons · start with identity without a central authority
12. computer sciencesimulation, evaluation, and complexity treated as engineering inputs.5 lessons · start with what minimal systems can show
13. correctnessthe methods that make agent-written software hold up: proofs, properties, replay, and receipts.15 lessons · start with unreasonably robust programming: a working definition