27 weeks · 54 posts · Written while building
Field notes from a personal AI OS in flight
Every Tuesday, an evergreen essay on what I'm learning while shipping DuranteOS. Every Friday, a dispatch from the week. Roughly 108,000 words and counting — for builders who'd rather watch the foundation get poured than read the press release.
Start here
If you're thinking about writing while building
After Twenty-Seven Weeks: What the Series Was For, and What I'm Building Next
If you're considering the founding cohort
The One Reference Customer Strategy: GTM for a Personal AI OS, Sketched Before the Customer Signs
If you want the architectural spine
The Decomposition Discipline I Am Trying to Codify Inside DOS
Subscribe · Tuesday essay
Around 3,800 builders read this weekly.
Latest dispatchW27 of 27
The Week Exclusivity Died: Substrate Goes Plural, Rules Layer Becomes the Moat
On Monday Microsoft and OpenAI signed an amended partnership ending the unilateral Azure exclusivity that had defined them since 2019. On Tuesday OpenAI's frontier models landed on AWS Bedrock, the same week Google took the Pentagon contract Anthropic refused on principle. On Wednesday Microsoft printed a $37B AI annual run-rate up 123%; on Thursday Fortune showed that ~$45B of "AI profits" at Alphabet and Amazon came from marking up Anthropic equity stakes; the same day Apple confirmed Gemini powers the next Siri. Three threads the series tracked all twenty-seven weeks resolve in the same five days. The series ends here. The substrate keeps moving.
W27 of 27
After Twenty-Seven Weeks: What the Series Was For, and What I'm Building Next
The series ends here. Twenty-seven weeks of evergreens, twenty-seven Friday dispatches, six months of writing while building. The last thirty-three weeks produced about three hundred and five commits, a Studio platform that has been live for one hundred and three days, a knowledge graph the writing forced me to articulate, and a list — drawn from the posts themselves — of what I am committing to build in the next quarter. This essay closes the arc and names the next one.
Weekly DispatchW26 of 27
The Substrate Signs a Ten-Year Lease: Gigawatts, Governance, and the Closing Window for Indie Posture
On Monday Amazon stapled $5B more into Anthropic alongside a $100B AWS commitment in return. On Wednesday Sundar Pichai opened Google Cloud Next 2026 with the reframe from "can we build an agent" to "how do we manage thousands," then shipped 8th-gen TPUs and a $750M partner fund. On Thursday OpenAI shipped GPT-5.5 with 40% fewer output tokens for the same Codex task. On Friday Google wrote Anthropic a $40B option at a $350B valuation against 5 GW of TPU capacity — and DeepSeek dropped V4 with 1M-token context as the open-weights default. The substrate stops being a service you rent and starts being a contract you sign, measured in years, gigawatts, and equity.
W26 of 27
The Substrate Portability Pact: What DOS Refuses to Couple to, Even as the Harness Consolidates
Studio has been live for ninety-nine days. Last Tuesday Anthropic shipped Claude Code Routines — saved configurations that run on Anthropic infrastructure, on Anthropic's schedule, behind Anthropic's auth. The harness layer just consolidated into a first-party product. This essay is the architectural commitment I am writing in public, while the temptation to couple is fresh — three things DOS will refuse to bind to even when refusing costs me speed, and the test that catches the violation.
Weekly DispatchW25 of 27
Routines Eat the Workflow: The Harness Becomes the Product
On Tuesday Anthropic shipped Claude Code Routines — saved Claude Code configs that run on Anthropic's cloud infrastructure on schedules, GitHub events, or POST-with-bearer-token webhooks. Pro five per day, Max fifteen, Team and Enterprise twenty-five. The desktop app was rebuilt around parallel sessions and an integrated terminal. The same week, OpenAI acqui-hired Hiro Finance to bolt vertical AI into ChatGPT, Jane Street signed a $6B AI-cloud agreement with CoreWeave, Alibaba shipped Qwen3.6-35B-A3B open-weights for consumer hardware, and the Anthropic Mythos / Project Glasswing partner roster expanded with bug findings: a 17-year FreeBSD RCE, less than 1% patched. The harness layer above the model is consolidating, and indie founders need to know whose harness orchestrates their workflows next quarter.
W25 of 27
The 90-Day Open-Weights Bakeoff: What Actually Routes Where in DOS Today
Studio has been routing across multiple substrates for ninety-five days. Seven providers integrated, ranging from Claude Sonnet 4.6 at one extreme to Google's open-weights Gemma 4 at the other. This essay is the empirical retrospective — which model handles which call type, where the routing logic sends each request by default, where the fallback kicks in, what the actual cost-per-quality numbers look like across roughly thirty thousand routed calls. Kent Beck on the smallest-experiment discipline applied to substrate selection. Andy Hunt and Dave Thomas on the Knowledge Portfolio reading: diversify, review, rebalance. Both applied to the moment routing decisions stop being intuition and become data.
Weekly DispatchW24 of 27
Project Glasswing: When the Frontier Bifurcates on Safety
On Tuesday Anthropic shipped Claude Mythos Preview — but not as an API. Under 'Project Glasswing,' Mythos went to a closed cohort of seven defenders: Amazon, Apple, Cisco, CrowdStrike, Linux Foundation, Microsoft, Palo Alto Networks. $100M in usage credits. $4M to open-source security organizations. Mythos found a 27-year-old OpenBSD bug. A 16-year-old FFmpeg bug. Autonomous Linux-kernel exploit chains. 181 working Firefox exploits versus Opus 4.6's 2. The model is gated because it is too capable. Meta meanwhile shipped Muse Spark to consumers. OpenAI's ad-revenue projections leaked at $100B by 2030. Claude Code quietly shipped a Monitor primitive that deletes whole categories of polling glue code. Five days, three theories of access, one bifurcation along a safety axis.
W24 of 27
Forking MemPalace: The 48-Hour Integration Retro
MemPalace shipped open-source on April 5. By Tuesday morning I had forked it into DOS. This is the integration retrospective written within forty-eight hours of the upstream launch — what I kept from my December sketch, what I had to abandon, what surprised me about the upstream's design choices, and what the operator-facing behavior actually looks like once a real implementation runs in production. Eric Evans on whether Wings/Rooms/Drawers map to Bounded Contexts (mostly yes). Greg Young on whether verbatim append-only storage is event-sourcing in the strict sense (mostly yes, with one important caveat I did not anticipate).
Weekly DispatchW23 of 27
The Agent Stack Swallows the IDE, the IPO, and the Courtroom
On Thursday Cursor 3 replaced the Composer pane with a parallel Agents Window, declaring the 'third era' where developers orchestrate agents rather than type. The same day, Google open-sourced Gemma 4 with the 31B Dense variant ranking #3 on Arena's open leaderboard. Microsoft shipped Copilot Cowork on Monday with both Claude and GPT cooperating inside one product. Q1 closed at roughly $300B in startup funding with AI taking 80% — capital chose its side. SpaceX-xAI filed a confidential S-1 targeting a $1.75T valuation. The DOJ appealed Judge Lin's injunction to the Ninth Circuit. Five days, five layers of the AI stack rebuilt around agents in parallel.
W23 of 27
Sentinel's First Scan: What the Convention Catalog Found Across Eleven Projects
Sentinel shipped in late March, fourteen weeks after I sketched it in W8. The first weekend after launch, I ran it across all eleven projects in DOS's project registry. This is the post-launch retrospective — what the convention catalog actually surfaced, where the scanner got it right on first pass, where it found patterns I had not seen in twenty-eight weeks of working in those repos, and where the scan was wrong in ways I did not expect. Sandi Metz on the squint test as the unit of pattern recognition. Michael Feathers on characterization tests as the unit of capture. Both applied to the moment a scanner reads a codebase as the domain expert.
Weekly DispatchW22 of 27
The Injunction the Pentagon Won't Honor: Anthropic Wins, the State Defies
On Thursday Judge Rita Lin granted Anthropic a preliminary injunction against the federal-procurement supply-chain-risk designation, calling it 'classic illegal First Amendment retaliation.' Hours later, Pentagon CTO Emil Michael declared on X that the order contained 'dozens of factual errors' and that the designation remained 'in full force and effect' — open defiance of a federal injunction. The same week, Claude Computer Use shipped to Pro/Max users, Anthropic's unannounced 'Mythos' next-tier model leaked via a CMS misconfiguration, and Shield AI raised $1.5B at a $12.7B valuation on the back of a US Air Force deal. Five days, one signal: the procurement story is now a constitutional one, and the architecture I committed to in Tuesday's evergreen just got its first stress test.
W22 of 27
Routing by Sovereignty Class: The Architecture That Survives the Procurement Decade
Studio's gateway started the year routing on capability and cost. By March it routes on three additional axes — sovereignty class, partner channel, and policy posture — because the procurement decisions of the last six weeks made all three load-bearing for any product whose customers might be evaluated against federal, EU, or sector-tier procurement frameworks. This is the architecture I want to commit to before the next sovereign decision lands. Cockburn's hexagonal lens. Greg Young's event-sourced audit trail. Both applied to the moment routing decisions stop being optimization and become governance.
Weekly DispatchW21 of 27
Cursor Becomes a Model Lab: Composer 2 and the Unbundling at the Agent Layer
On Thursday Cursor shipped Composer 2 as its default agentic coding model — and within twenty-four hours sleuths confirmed via API headers that it is a fine-tune of Kimi K2.5. The IDE became a model lab in nine months. OpenAI made GPT-5.4 mini free on Tuesday. The DOD escalated the Anthropic case to 'unacceptable risk to national security' Wednesday, with a preliminary-injunction hearing fast-tracked to next week. Anthropic published an 81,000-person user study the same day. Perplexity made the $200/month Comet browser free on iPhone. The frontier is unbundling at the agent layer. The coding-tool moat just became a training run, not a wrapper.
W21 of 27
MCP in Production: What Works When the Protocol Becomes the Boundary
Studio has been routing MCP-shaped tool calls in production for sixty-seven days. The protocol holds where I expected it to and leaks where I did not. This is the evergreen I want on the record before the next quarter of building, with Cockburn's hexagonal lens and Fowler's anti-corruption layer holding the framing. What MCP buys you in production, what it does not, and the three places I have already had to write provider-specific code despite the protocol's promise.
Weekly DispatchW20 of 27
The Counter-Sovereignty Move: Anthropic Sues, Then Diversifies
On Monday Anthropic sued the US Department of War in N.D. Cal. and the D.C. Circuit, challenging last week's federal-procurement supply-chain-risk designation under five separate counts including First Amendment retaliation. The same week the company opened its fourth APAC office in Sydney, launched the Anthropic Institute as its public-benefit research arm with Jack Clark at the helm, and committed $100M to a Claude Partner Network with Accenture, Deloitte, Cognizant, and Infosys. OpenAI quietly deprecated the GPT-5.1 family alongside. Five days, one signal: when a sovereign cuts you off, you spread the surface area — and you do it faster than the news cycle can catch you in any single position.
W20 of 27
Six Months of DOS: What Changed, What Endured, What Surprised
Twenty-six weeks of building a personal AI operating system in public. Twenty essays in this series so far. Roughly twelve packs in various states of readiness. Around eighty workflows. About 3,800 newsletter subscribers. Zero paying customers, with four months left on the founding-customer clock. Here is the honest half-year retrospective — what survived contact with reality through Q1, what I was wrong about, what I never saw coming.
Weekly DispatchW19 of 27
Procurement Is the New Benchmark: OpenAI Wins the DoW While Anthropic Gets Banned
On Monday OpenAI signed a Department of War classified-systems agreement — the same day the administration designated Anthropic a federal-procurement supply-chain risk. Tuesday brought Google's Gemini 3.1 Flash-Lite. Wednesday-Thursday: Anthropic temporarily downgraded Claude Code's default reasoning effort and Dario Amodei published a public response to the DoW letter. Thursday brought GPT-5.4 with a 1M-token context, Tool Search, and record OSWorld + WebArena scores. Cursor shipped Automations the same day with event-triggered agents. Five days, one signal: procurement is the new benchmark, and indie founders building on a single-vendor spine just had the sovereignty risk priced in public for the first time.
W19 of 27
The Knowledge Portfolio for an Indie Founder Building in Public, Audited at Week 25
Andy Hunt and Dave Thomas argued in The Pragmatic Programmer that engineers should manage their learning the way an investor manages a portfolio — diversify, invest regularly, review periodically, accept calculated risk. Twenty-six years later, applied honestly twenty-five weeks into building a personal AI OS, here is what my actual portfolio looks like, where the binding constraint is, and what I am rebalancing this quarter.
Weekly DispatchW18 of 27
The Moat Above the Model: Distillation, Mobile Coding, and MCP Battlegrounds
On Monday Anthropic accused DeepSeek, Moonshot, and MiniMax of industrial-scale Claude distillation — twenty-four thousand fake accounts, sixteen million exchanges, the moat narrative reframed in a single press cycle. Tuesday brought Claude Code Remote Control plus the disclosure that Claude Code now runs at $2.5B ARR. Wednesday: OpenAI announced the Frontier Alliance with McKinsey, BCG, Accenture, and Capgemini. Thursday: Figma shipped Codex roundtrip integration via MCP, exactly one week after its parallel Claude Code integration. Friday rounded out with Anthropic's acquisition of Vercept for computer-use. Five days, four labs, one signal: the value has decisively moved above the model. The model is now substrate; the agent stack is the moat.
W18 of 27
Skills, Packs, and Hooks: The Three-Layer Model I'm Pulling DOS Toward
Anthropic shipped Skills as an open standard in December. The model I want DOS to settle on adds two more layers above and around it — Packs as Bounded-Context bundles of related Skills, and Hooks as the lifecycle layer that fires without invocation. Twenty-four weeks into DOS, with Studio shipping forty-six days ago and the open Skills spec now table stakes across coding-agent surfaces, this is the three-layer architecture I want to commit to before the layering decisions get baked the wrong way.
The 27-week arc · A single body of work
Twenty-seven weeks. Two posts a week. Six months of writing while building.
Week
Tuesday evergreen
Friday dispatch
2025-10-28
The Agent Is the Product: Why Intelligence Replaces Interface
2025-10-31
Wrapper Wars: The Week Cursor, Windsurf, and MiniMax All Went Vertical
2025-11-04
Memory Replaces Lock-In: Designing the Substrate of Personal AI
2025-11-07
The Week the Substrate Caught Up: Kimi K2 Thinking Lands
2025-11-11
Why I Just Left a Steady Practice to Build a Personal AI Operating System
2025-11-14
The Week the Harness Ate the Model
2025-11-18
The Decomposition Discipline I Am Trying to Codify Inside DOS
2025-11-21
Antigravity, Gemini 3, Grok 4.1: The Week the Frontier Re-Rendered
2025-11-25
Four Copies, One Source of Truth: The Sync Pattern I Want to Commit to Before It Hurts
2025-11-28
Opus 4.5 Lands on Thanksgiving Week: Anthropic's Quiet Counter-Punch
2025-12-02
Plugin Architecture for Hooks: The Pattern I Want Before the Hook Becomes a God-Function
2025-12-05
The Runtime Is the Moat Now: Anthropic Buys Bun, Claude Code Hits a Billion
2025-12-09
Credit-Metered API Gateways: The Pattern I Want Before I Have Anything to Meter
2025-12-12
Three Frontier Models in 23 Days: Stop Picking a Winner, Start Picking a Router
2025-12-16
Sentinel, Sketched: Convention-Driven Onboarding Before I Build It
2025-12-19
The Spec Goes Public, the Substrate Goes to War: Skills Opens While Codex, Gemini Flash, and Nemotron Sprint
2025-12-23
Memory, Sketched: The Knowledge Graph I Was Designing Before MemPalace Shipped
2025-12-26
Christmas Wasn't Quiet: Open Weights Caught Up, Nvidia Bought Inference, Frontier Labs Bought Loyalty
2025-12-30
Council, Sketched: The Three-Round Debate Protocol I Want to Formalize Before I Build It
2026-01-02
The Practitioners' Year-End: While the West Took PTO, DeepSeek Shipped Architecture
2026-01-06
Altyaa's Wedge: The Brazilian-Portuguese SMB Bet Studio Is Pointed At
2026-01-09
Default Claude: Microsoft Flips the Switch as the Substrate Distributes by Default
2026-01-13
Builder's Compass: Two Years In, ~3,800 Subscribers, and What I've Learned Teaching Architecture
2026-01-16
The Agent Surface Splits in Two: Anthropic Goes Vertical While China Goes Substrate
2026-01-20
Failure Patterns, Sketched: The Bookkeeping Discipline I Want for the Agent's Mistakes
2026-01-23
The Rules Layer Solidifies: Constitution, Hardware, and Export Controls Land in One Week
2026-01-27
The One Reference Customer Strategy: GTM for a Personal AI OS, Sketched Before the Customer Signs
2026-01-30
Vertical Integration Eats Horizontal AI: Maia, Meta's Capex, and Anthropic-ServiceNow
2026-02-03
Hexagonal in Practice: The Ports and Adapters I'm Pulling Studio Toward
2026-02-06
Coding Becomes the Flagship: Opus 4.6 and GPT-5.3-Codex Land the Same Day
2026-02-10
TDD for AI Agents, Sketched: The Translation I Want to Commit to Before the Eval Suite Exists
2026-02-13
The Bifurcation Hardens: Anthropic's $380B Meets China's Open-Weights Frontier
2026-02-17
Refactoring the Hook Pipeline: The Fowler Walkthrough I'm Three Refactorings Into
2026-02-20
The Sonnet Window: Free-Tier Skills, Code Security, and the Race for Agentic Primitives
2026-02-24
Skills, Packs, and Hooks: The Three-Layer Model I'm Pulling DOS Toward
2026-02-27
The Moat Above the Model: Distillation, Mobile Coding, and MCP Battlegrounds
2026-03-03
The Knowledge Portfolio for an Indie Founder Building in Public, Audited at Week 25
2026-03-06
Procurement Is the New Benchmark: OpenAI Wins the DoW While Anthropic Gets Banned
2026-03-10
Six Months of DOS: What Changed, What Endured, What Surprised
2026-03-13
The Counter-Sovereignty Move: Anthropic Sues, Then Diversifies
2026-03-17
MCP in Production: What Works When the Protocol Becomes the Boundary
2026-03-20
Cursor Becomes a Model Lab: Composer 2 and the Unbundling at the Agent Layer
2026-03-24
Routing by Sovereignty Class: The Architecture That Survives the Procurement Decade
2026-03-27
The Injunction the Pentagon Won't Honor: Anthropic Wins, the State Defies
2026-03-31
Sentinel's First Scan: What the Convention Catalog Found Across Eleven Projects
2026-04-03
The Agent Stack Swallows the IDE, the IPO, and the Courtroom
2026-04-07
Forking MemPalace: The 48-Hour Integration Retro
2026-04-10
Project Glasswing: When the Frontier Bifurcates on Safety
2026-04-14
The 90-Day Open-Weights Bakeoff: What Actually Routes Where in DOS Today
2026-04-17
Routines Eat the Workflow: The Harness Becomes the Product