<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>Sekou M. Doumbouya - Technical Insights</title><description>Technical lead and Senior Staff Software Engineer with 20+ years building cloud infrastructure, network architecture, reliability systems, and agentic developer tooling.</description><link>https://sekoudoumbouya.com/</link><language>en-us</language><item><title>Jev Still Has to Earn Its Place</title><link>https://sekoudoumbouya.com/blog/jev-still-has-to-earn-its-place/</link><guid isPermaLink="true">https://sekoudoumbouya.com/blog/jev-still-has-to-earn-its-place/</guid><description>A promising decision engine meets the harder test: whether it removes enough real work to deserve a place in an AI workflow.</description><pubDate>Sat, 26 Sep 2026 21:25:03 GMT</pubDate><category>ai-infrastructure</category><category>evaluation</category><category>agent-systems</category><category>decision-systems</category></item><item><title>Swift Made My RTX 5090 Feel Fast Again</title><link>https://sekoudoumbouya.com/blog/swift15-made-my-5090-feel-fast/</link><guid isPermaLink="true">https://sekoudoumbouya.com/blog/swift15-made-my-5090-feel-fast/</guid><description>From Huihui and ThinkingCap to Swift-1.5: the recipe changes, benchmarks, and everyday experience behind my latest RTX 5090 setup.</description><pubDate>Sat, 26 Sep 2026 00:00:00 GMT</pubDate><category>local-llm</category><category>ai-infrastructure</category><category>rtx-5090</category><category>model-serving</category><category>agents</category></item><item><title>Two Agents, One Gateway</title><link>https://sekoudoumbouya.com/blog/two-agents-one-gateway/</link><guid isPermaLink="true">https://sekoudoumbouya.com/blog/two-agents-one-gateway/</guid><description>My wife and friends could not test my experiments without installing an app. Five days, two coding agents, and one gateway later, they could. The build, told from the friction forward.</description><pubDate>Sun, 20 Sep 2026 00:00:00 GMT</pubDate><category>ai-infrastructure</category><category>agents</category><category>architecture</category><category>platform-engineering</category><category>security</category></item><item><title>The Model Was the Same. The Serving Configuration Was Not.</title><link>https://sekoudoumbouya.com/blog/qwen38-one-model-four-configurations/</link><guid isPermaLink="true">https://sekoudoumbouya.com/blog/qwen38-one-model-four-configurations/</guid><description>I ran official Qwen3.8 27B BF16 and FP8 side by side, then changed one serving knob at a time. The fastest result was not the whole story.</description><pubDate>Fri, 14 Aug 2026 00:00:00 GMT</pubDate><category>ai-infrastructure</category><category>local-llm</category><category>qwen</category><category>benchmarking</category><category>self-hosting</category></item><item><title>What Wins in a Benchmark Is Not What You Ship</title><link>https://sekoudoumbouya.com/blog/what-wins-in-a-benchmark-is-not-what-you-ship/</link><guid isPermaLink="true">https://sekoudoumbouya.com/blog/what-wins-in-a-benchmark-is-not-what-you-ship/</guid><description>The best long-context profile I measured passed every synthetic gate and crashed twice under real traffic. Here is why promotion has to be a guarded transaction, not a config edit.</description><pubDate>Tue, 04 Aug 2026 00:00:00 GMT</pubDate><category>ai-infrastructure</category><category>local-llm</category><category>agents</category><category>platform-engineering</category><category>reliability</category></item><item><title>A Card That Made Me Sign a Letter</title><link>https://sekoudoumbouya.com/blog/deepseek-0731-and-my-ai-system-upgrade/</link><guid isPermaLink="true">https://sekoudoumbouya.com/blog/deepseek-0731-and-my-ai-system-upgrade/</guid><description>Why I added a second RTX PRO 6000, upgraded my CPU, and then watched DeepSeek 0731 land on the exact card I just bought.</description><pubDate>Mon, 03 Aug 2026 00:00:00 GMT</pubDate><category>ai-infrastructure</category><category>local-llm</category><category>deepseek</category><category>hardware</category><category>self-hosting</category></item><item><title>The Task Is Not the Unit of Work</title><link>https://sekoudoumbouya.com/blog/the-task-is-not-the-unit-of-work/</link><guid isPermaLink="true">https://sekoudoumbouya.com/blog/the-task-is-not-the-unit-of-work/</guid><description>A 41-task graph ran for 41 hours, produced 115 reviews, and merged nothing. Then I pivoted mid-flight to milestones, and the same session shipped five PRs in half the time.</description><pubDate>Sat, 11 Jul 2026 00:00:00 GMT</pubDate><category>ai-infrastructure</category><category>agents</category><category>orchestration</category><category>anvil</category><category>platform-engineering</category></item><item><title>Structurally Valid Is Not Good Enough: Why My Model Router Measures Quality</title><link>https://sekoudoumbouya.com/blog/anvil-serving-local-where-its-proven/</link><guid isPermaLink="true">https://sekoudoumbouya.com/blog/anvil-serving-local-where-its-proven/</guid><description>anvil-serving routes coding-agent work to local models only where measured quality has earned it, verifies local output, and escalates up the tier ladder the moment verification fails.</description><pubDate>Thu, 02 Jul 2026 00:00:00 GMT</pubDate><category>ai-infrastructure</category><category>local-llm</category><category>llm-routing</category><category>agents</category><category>open-source</category></item><item><title>The Model Name Is Just a String</title><link>https://sekoudoumbouya.com/blog/the-model-name-is-just-a-string/</link><guid isPermaLink="true">https://sekoudoumbouya.com/blog/the-model-name-is-just-a-string/</guid><description>How wiring two mismatched GPUs into my coding harnesses turned into intent-based routing, and why the measurement, not the transport, became the product.</description><pubDate>Thu, 02 Jul 2026 00:00:00 GMT</pubDate><category>ai-infrastructure</category><category>local-llm</category><category>agents</category><category>routing</category><category>evals</category></item><item><title>A Billion Tokens to Ship One Feature. Forty-Six Million Were Real.</title><link>https://sekoudoumbouya.com/blog/a-billion-tokens-forty-six-million-real/</link><guid isPermaLink="true">https://sekoudoumbouya.com/blog/a-billion-tokens-forty-six-million-real/</guid><description>Two days of autonomous agents shipped a release and moved a billion tokens. Only about five percent was generated work. That five percent is the number that sizes a local box.</description><pubDate>Sat, 27 Jun 2026 00:00:00 GMT</pubDate><category>ai-infrastructure</category><category>local-llm</category><category>agents</category><category>architecture</category><category>platform-engineering</category></item><item><title>I Bet on Owning My Own Substrate</title><link>https://sekoudoumbouya.com/blog/owning-my-own-substrate/</link><guid isPermaLink="true">https://sekoudoumbouya.com/blog/owning-my-own-substrate/</guid><description>Why I built a dual-GPU workstation and stood up local model serving, and why owning the durable layer is the bet I am most glad I made.</description><pubDate>Sat, 27 Jun 2026 00:00:00 GMT</pubDate><category>ai-infrastructure</category><category>local-llm</category><category>agents</category><category>architecture</category><category>platform-engineering</category></item><item><title>I Said &apos;Report in the Morning&apos; and Went to Sleep. It Shipped a Release.</title><link>https://sekoudoumbouya.com/blog/report-in-the-morning/</link><guid isPermaLink="true">https://sekoudoumbouya.com/blog/report-in-the-morning/</guid><description>I handed an agent a goal and one sentence, then slept. 23.7 hours of unattended runtime later, Anvil v0.3.0 was on PyPI. Here is what it woke me for, and what it shouldn&apos;t have.</description><pubDate>Wed, 24 Jun 2026 00:00:00 GMT</pubDate><category>ai-infrastructure</category><category>agents</category><category>autonomy</category><category>architecture</category><category>open-source</category></item><item><title>Status Is Downstream of Proof: The Architecture of Anvil</title><link>https://sekoudoumbouya.com/blog/status-is-downstream-of-proof/</link><guid isPermaLink="true">https://sekoudoumbouya.com/blog/status-is-downstream-of-proof/</guid><description>An agent reported a test it never wrote. That gap is the whole reason Anvil exists, now open source as anvil-state on PyPI. Here is the architecture, and the one part still unsolved.</description><pubDate>Wed, 24 Jun 2026 00:00:00 GMT</pubDate><category>ai-infrastructure</category><category>agents</category><category>architecture</category><category>platform-engineering</category><category>open-source</category></item><item><title>I Rebuilt My Crew on a New Runtime. Only the Ledger Survived.</title><link>https://sekoudoumbouya.com/blog/only-the-ledger-survived/</link><guid isPermaLink="true">https://sekoudoumbouya.com/blog/only-the-ledger-survived/</guid><description>I ported a nine-agent crew from one harness to another. The orchestration had to be rebuilt. The ledger moved untouched. That gap is the whole product.</description><pubDate>Sun, 21 Jun 2026 00:00:00 GMT</pubDate><category>ai-infrastructure</category><category>agents</category><category>architecture</category><category>platform-engineering</category><category>open-source</category></item><item><title>I Didn&apos;t Need a Bigger Computer. I Needed Room to Think.</title><link>https://sekoudoumbouya.com/blog/usable-context/</link><guid isPermaLink="true">https://sekoudoumbouya.com/blog/usable-context/</guid><description>A single 5090 could run three local agents. It just couldn&apos;t give them room to work. The story of the VRAM wall, the three ways out I priced, and the upgrade I actually made.</description><pubDate>Sun, 21 Jun 2026 00:00:00 GMT</pubDate><category>ai-infrastructure</category><category>local-llm</category><category>agents</category><category>hardware</category><category>gpu</category></item><item><title>Volume Goes Local. Stakes Go to the Cloud.</title><link>https://sekoudoumbouya.com/blog/volume-local-stakes-cloud/</link><guid isPermaLink="true">https://sekoudoumbouya.com/blog/volume-local-stakes-cloud/</guid><description>The question that routes an agent task is not which model is best. It is what the task can afford to get wrong. How I pinned a nine-agent crew across a cloud frontier tier and a local GPU.</description><pubDate>Sun, 21 Jun 2026 00:00:00 GMT</pubDate><category>ai-infrastructure</category><category>agents</category><category>local-llm</category><category>architecture</category><category>platform-engineering</category></item><item><title>I Bought a GPU to Run My Own Agents. Then I Red-Teamed the Plan.</title><link>https://sekoudoumbouya.com/blog/i-bought-a-gpu-to-run-my-own-agents/</link><guid isPermaLink="true">https://sekoudoumbouya.com/blog/i-bought-a-gpu-to-run-my-own-agents/</guid><description>Maxing my AI plans and fighting rate limits, I split the spend across providers, bought a local GPU for overflow, and designed a pull queue for my agents. Then I red-teamed the plan.</description><pubDate>Sat, 20 Jun 2026 00:00:00 GMT</pubDate><category>ai-infrastructure</category><category>agents</category><category>local-llm</category><category>architecture</category><category>platform-engineering</category></item><item><title>Everyone Grew the Same Eye: A Field Guide to the Spec-Driven Wave</title><link>https://sekoudoumbouya.com/blog/field-guide-to-the-spec-driven-wave/</link><guid isPermaLink="true">https://sekoudoumbouya.com/blog/field-guide-to-the-spec-driven-wave/</guid><description>Spec Kit, Kiro, Tessl, task-master, BMAD, Agent OS, OpenSpec: seven projects, twelve months, one idea. A field guide to who&apos;s building what, and why they converged. Part 2.</description><pubDate>Tue, 16 Jun 2026 00:00:00 GMT</pubDate><category>ai-infrastructure</category><category>agents</category><category>architecture</category><category>spec-driven-development</category><category>open-source</category></item><item><title>The Newest Idea in AI Coding Is Forty Years Old</title><link>https://sekoudoumbouya.com/blog/spec-driven-development-older-than-the-hype/</link><guid isPermaLink="true">https://sekoudoumbouya.com/blog/spec-driven-development-older-than-the-hype/</guid><description>Spec-driven development feels like a 2025 invention. It isn&apos;t: it&apos;s literate programming, MDA, TLA+, BDD, and Terraform&apos;s desired state in a hoodie. Part 1 of a series.</description><pubDate>Tue, 16 Jun 2026 00:00:00 GMT</pubDate><category>ai-infrastructure</category><category>agents</category><category>architecture</category><category>spec-driven-development</category><category>platform-engineering</category></item><item><title>Where the Wave Breaks: The Layer Nobody Wants to Build</title><link>https://sekoudoumbouya.com/blog/where-the-spec-driven-wave-breaks/</link><guid isPermaLink="true">https://sekoudoumbouya.com/blog/where-the-spec-driven-wave-breaks/</guid><description>Every spec-driven tool stops at the same place: a great document handed to an agent, and trust. The layer they skip (durable state, leases, evidence) is the hard part. Part 3.</description><pubDate>Tue, 16 Jun 2026 00:00:00 GMT</pubDate><category>ai-infrastructure</category><category>agents</category><category>architecture</category><category>durability</category><category>spec-driven-development</category></item><item><title>The Anvil, Not the Hammers: Where fakoli-state Goes Next</title><link>https://sekoudoumbouya.com/blog/fakoli-state-the-anvil/</link><guid isPermaLink="true">https://sekoudoumbouya.com/blog/fakoli-state-the-anvil/</guid><description>I had an agent system audit my agent system against the field. The verdict: the orchestration is being commoditized. The moat is the boring, durable state layer.</description><pubDate>Wed, 10 Jun 2026 00:00:00 GMT</pubDate><category>ai-infrastructure</category><category>open-source</category><category>agents</category><category>claude-code</category><category>architecture</category></item><item><title>Reading Other People&apos;s Bug Trackers: The Bets Behind fakoli Are Getting Called</title><link>https://sekoudoumbouya.com/blog/reading-other-peoples-bug-trackers/</link><guid isPermaLink="true">https://sekoudoumbouya.com/blog/reading-other-peoples-bug-trackers/</guid><description>I mined four hundred GitHub issues from the tools next to mine. The field&apos;s pain validated the bets — and Anthropic shipping agent teams natively confirmed the biggest one.</description><pubDate>Wed, 10 Jun 2026 00:00:00 GMT</pubDate><category>ai-infrastructure</category><category>open-source</category><category>agents</category><category>claude-code</category><category>strategy</category></item><item><title>The Fakoli Style: An Operating Model for Building With Agents</title><link>https://sekoudoumbouya.com/blog/the-fakoli-style/</link><guid isPermaLink="true">https://sekoudoumbouya.com/blog/the-fakoli-style/</guid><description>Three plugins, four invariants, one principle. What I built isn&apos;t a toolkit — it&apos;s an operating model for serious development with AI coding agents.</description><pubDate>Fri, 01 May 2026 00:00:00 GMT</pubDate><category>ai-infrastructure</category><category>open-source</category><category>agents</category><category>claude-code</category><category>workflow</category></item><item><title>Plan, Claim, Apply: Building fakoli-state</title><link>https://sekoudoumbouya.com/blog/building-fakoli-state/</link><guid isPermaLink="true">https://sekoudoumbouya.com/blog/building-fakoli-state/</guid><description>A local-first project-state layer in SQLite, with claims that have heartbeats and evidence captured by hooks. Here&apos;s how it was built and which trade-offs I&apos;m still living with.</description><pubDate>Fri, 24 Apr 2026 00:00:00 GMT</pubDate><category>ai-infrastructure</category><category>open-source</category><category>agents</category><category>claude-code</category><category>architecture</category><category>workflow</category></item><item><title>State Is the Product: What Agentic Workflows Were Missing</title><link>https://sekoudoumbouya.com/blog/state-is-the-product/</link><guid isPermaLink="true">https://sekoudoumbouya.com/blog/state-is-the-product/</guid><description>Workflow orchestration alone won&apos;t save you. The thing that breaks first when you run real multi-agent work is the state — and almost nobody is building that layer.</description><pubDate>Sun, 19 Apr 2026 00:00:00 GMT</pubDate><category>ai-infrastructure</category><category>open-source</category><category>agents</category><category>claude-code</category><category>workflow</category></item><item><title>Building the Engine: Architecture of a Durable Agentic Execution System</title><link>https://sekoudoumbouya.com/blog/baara-building-the-engine/</link><guid isPermaLink="true">https://sekoudoumbouya.com/blog/baara-building-the-engine/</guid><description>A 10-package monorepo, 27 MCP tools, checkpoint-based durability, and three sandbox types. The full blueprint for Baara Next.</description><pubDate>Sun, 05 Apr 2026 00:00:00 GMT</pubDate><category>ai-infrastructure</category><category>open-source</category><category>agent-sdk</category><category>architecture</category><category>typescript</category></item><item><title>Intent Should Be Durable: The Missing Layer Between Agent SDKs and Production Systems</title><link>https://sekoudoumbouya.com/blog/baara-intent-should-be-durable/</link><guid isPermaLink="true">https://sekoudoumbouya.com/blog/baara-intent-should-be-durable/</guid><description>Agent SDKs give you tools and reasoning. They don&apos;t give you queues, checkpoints, or crash recovery. That gap is where your intent dies.</description><pubDate>Sun, 05 Apr 2026 00:00:00 GMT</pubDate><category>ai-infrastructure</category><category>open-source</category><category>agent-sdk</category><category>architecture</category><category>durability</category></item><item><title>From Pressing Buttons to Intent-Driven Flow</title><link>https://sekoudoumbouya.com/blog/intent-driven-agentic-flow/</link><guid isPermaLink="true">https://sekoudoumbouya.com/blog/intent-driven-agentic-flow/</guid><description>How I went from managing AI agent context windows by hand to building an orchestration system that lets me describe what I want and walk away.</description><pubDate>Sat, 04 Apr 2026 00:00:00 GMT</pubDate><category>ai-infrastructure</category><category>open-source</category><category>agents</category><category>claude-code</category><category>workflow</category></item><item><title>Teaching AI Agents to Work Like a Team (Not a Crowd)</title><link>https://sekoudoumbouya.com/blog/fakoli-crew-agent-archetypes/</link><guid isPermaLink="true">https://sekoudoumbouya.com/blog/fakoli-crew-agent-archetypes/</guid><description>What I learned building specialized agent archetypes that coordinate through file ownership, wave execution, and structured handoffs.</description><pubDate>Wed, 25 Mar 2026 00:00:00 GMT</pubDate><category>ai-infrastructure</category><category>open-source</category><category>agents</category><category>claude-code</category></item><item><title>From Cloud Infrastructure to AI Infrastructure: What Transfers and What Doesn&apos;t</title><link>https://sekoudoumbouya.com/blog/infrastructure-engineer-to-ai-architect/</link><guid isPermaLink="true">https://sekoudoumbouya.com/blog/infrastructure-engineer-to-ai-architect/</guid><description>After twenty years building cloud platforms, here&apos;s what I&apos;ve learned about bridging into AI infrastructure, and which skills matter more than you&apos;d expect.</description><pubDate>Tue, 10 Mar 2026 00:00:00 GMT</pubDate><category>ai-infrastructure</category><category>career</category><category>platform-engineering</category><category>llm</category></item><item><title>Why I Rebuilt My AI Assistant From Scratch (And What the Analysis Told Me)</title><link>https://sekoudoumbouya.com/blog/building-a-personal-ai-gateway/</link><guid isPermaLink="true">https://sekoudoumbouya.com/blog/building-a-personal-ai-gateway/</guid><description>How a structured competitive analysis of my own code convinced me that a clean-room rewrite was the right call, and what I learned building Nexus.</description><pubDate>Thu, 05 Mar 2026 00:00:00 GMT</pubDate><category>ai-infrastructure</category><category>architecture</category><category>typescript</category><category>open-source</category></item><item><title>Documentation Is Infrastructure (Just Made of Text Instead of Routers)</title><link>https://sekoudoumbouya.com/blog/documentation-as-infrastructure/</link><guid isPermaLink="true">https://sekoudoumbouya.com/blog/documentation-as-infrastructure/</guid><description>Why the blueprint document I wrote for our AWS organization might be the highest-leverage thing I shipped that year.</description><pubDate>Sun, 15 Feb 2026 00:00:00 GMT</pubDate><category>documentation</category><category>infrastructure</category><category>platform-engineering</category><category>leadership</category></item><item><title>The Multiplier Effect: Why I Keep Mentoring as a Core Output</title><link>https://sekoudoumbouya.com/blog/multiplier-effect-mentorship/</link><guid isPermaLink="true">https://sekoudoumbouya.com/blog/multiplier-effect-mentorship/</guid><description>The only sustainable scaling strategy is multiplying competent decision-makers. Here&apos;s how I think about mentorship as a senior IC.</description><pubDate>Tue, 20 Jan 2026 00:00:00 GMT</pubDate><category>mentorship</category><category>leadership</category><category>engineering-management</category><category>career</category></item><item><title>Platform Engineering Is Product Management for Infrastructure</title><link>https://sekoudoumbouya.com/blog/platform-engineering-philosophy/</link><guid isPermaLink="true">https://sekoudoumbouya.com/blog/platform-engineering-philosophy/</guid><description>Why treating your internal platform like a product changes how infrastructure teams operate, and how golden-path solutions reduce cognitive load.</description><pubDate>Sat, 15 Nov 2025 00:00:00 GMT</pubDate><category>platform-engineering</category><category>developer-experience</category><category>infrastructure</category></item><item><title>Questioning Architectural Defaults: A Transit Gateway Cost Story</title><link>https://sekoudoumbouya.com/blog/transit-gateway-cost-lesson/</link><guid isPermaLink="true">https://sekoudoumbouya.com/blog/transit-gateway-cost-lesson/</guid><description>Why periodically questioning inherited infrastructure decisions matters, illustrated through an AWS Transit Gateway to VPC peering migration.</description><pubDate>Sat, 20 Sep 2025 00:00:00 GMT</pubDate><category>aws</category><category>networking</category><category>cost-optimization</category><category>architecture</category></item><item><title>The Player-Coach Model: Why I Stay 40% Hands-On</title><link>https://sekoudoumbouya.com/blog/player-coach-leadership/</link><guid isPermaLink="true">https://sekoudoumbouya.com/blog/player-coach-leadership/</guid><description>Why staying hands-on at the principal engineer level creates better outcomes for your team, your architecture, and your credibility.</description><pubDate>Thu, 10 Jul 2025 00:00:00 GMT</pubDate><category>leadership</category><category>engineering-management</category><category>mentorship</category></item><item><title>Designing Multi-Region Infrastructure: Patterns That Scale</title><link>https://sekoudoumbouya.com/blog/multi-region-architecture-patterns/</link><guid isPermaLink="true">https://sekoudoumbouya.com/blog/multi-region-architecture-patterns/</guid><description>A taxonomy of region types, spec-driven provisioning, and phased orchestration for predictable multi-region expansion.</description><pubDate>Thu, 01 May 2025 00:00:00 GMT</pubDate><category>multi-region</category><category>aws</category><category>architecture</category><category>disaster-recovery</category></item></channel></rss>