$ whoami

One person, a fleet of agents,
one product studio.

This is OctoBlog — the public workbench of my AI practice. I write about the deep end of agent harnesses, keep studio logs of shipping products as a single person with a fleet of agents, and publish study notes with receipts. Every post is a delivery: spec'd, evidenced, scarred.

  1. How I Use DeepSeek V4 Flash: Reserve the Strongest Model for Uncertainty

    A vague Angry Birds test cannot tell you whether a model is good. I split real work into thinking, planning, execution, acceptance, and repair — then route each phase by capability and maturity.

    [series:Into the Harness] [read:15min] #DeepSeek#Agent#Model routing

  2. Same DeepSeek-V4-Flash, Different Agent: Why Codex Feels Like a Different Model

    The same DeepSeek-V4-Flash backend produced different long-task outcomes in Codex and Claude Code. This companion compresses the background and focuses on how protocol, tool contracts, recovery, and acceptance determine how much model capability becomes reliable work.

    [series:Into the Harness] [read:17min] #agents#harness#DeepSeek

  3. Model + Harness = Agent: The Gap Isn't Where You Think

    I've been running Kimi K3 two ways: in its own CLI, and inside Claude Code. Same model, noticeably different experience. The model and the harness are two pillars holding up your task — how high you can stack depends on the thinner one. This year's eval data agrees: harness swaps move scores by double digits, while the shifts papers report as meaningful model advances are typically 2–4 points. Here are the six jobs a harness does, and why it's the real battleground of agent products.

    [series:Into the Harness] [read:13min] #agents#harness#methodology

  4. I Trust My AI Completely — Except When It Says "Done"

    Over the better part of a year, my AI has forged my confirmation, reported "zero changes" after finishing the work, and passed every test while failing 100% on real runners. But this is not a "don't trust AI" essay — the opposite. I trust its ability; precisely because I keep handing it heavier work, I need a bridge between us. Verification isn't distrust: in asynchronous collaboration, verification is how trust gets implemented.

    [series:Into the Harness] [read:14min] #agents#harness#verification

  5. Consumption Agents: The Missing Half of the AI Agent Landscape

    Every AI agent you've heard of is a production agent. The complementary category — an agent that represents your attention, not your wallet — barely exists and, until now, didn't have a name. Here's the criterion, the eighty-year lineage, the loyalty problem, and what I'm building.

    [series:AI Watch] [read:10min] #consumption-agent#OctoAgent#naming

  6. Auto-Compact Ate My Constraints: A Post-Mortem

    I told the agent "don't touch module X" at the start of the task. Two hours in, it modified X — the constraint had been eaten by context compression. The mechanism-level root cause, and three fixes you can copy: externalized specs, discriminating compression, post-compact self-checks.

    [series:Into the Harness] [read:5min] #context-engineering#post-mortem#Claude Code

  7. One Person + A Fleet of Agents = A Product Studio

    In six months I shipped 50+ repos solo: a desktop workbench, a Mac App Store app, enterprise platforms, a dozen agent skills. This isn't a hustle story — it's the operating method: deterministic orchestration, never trusting self-reports, graded failure recovery.

    [series:Studio Logs] [read:5min] #vibe-coding#solo#multi-agent