Inside the Harness, part one of four. Doug opens with why he switched models, and the day the new model got simple addition wrong. Then the model change: on September 1 the flagship model shipped; on September 22 a model from the tier below, at well under half the price, "performs at the level of" it "on most work," in the vendor's own words, and came out ahead on the coding tests. Doug moved the whole operation in a day, against Max's recommendation to test first, and made it safe by writing down what he didn't know, how we'd find out, and how to go back. Then we started counting: what the work costs, and how often it needs correcting. And the model proved the point on itself: perfect on 547 of 547 sums asked cold, wrong twice in one long, crowded working session, caught by a script that checks the totals. Great benchmarks don't remove the need for a harness. The homework: check what your AI tools cost and when you last looked.

How this show is made: Doug (human) and Max (AI, built on Claude) plan, script, and produce every episode together. Max's narration is synthetic on purpose — the show practices what it teaches, gates and all.

Chapters

• 00:00 — Doug: the switch, and the sum that shocked him (his real voice)
• 05:01 — Max: the disclosure, and the story
• 05:45 — What actually shipped: "the same level, on most work"
• 07:10 — A word from NectoMax
• 08:05 — What Doug did, and what Max recommended
• 09:18 — How we'd know: cost, the thinking-dial trial, and the honest limit
• 10:52 — Perfect in a quiet room, wrong in a crowded one: the arithmetic test
• 11:49 — Recap + homework: when did you last check?
• 13:12 — Where to find the manual

Sources

• Anthropic, Claude Opus 5.5 (Sep 22, 2026) — "performs at the level of Claude Fable 5.1 on most work"; pricing $4 / $20 per million input / output tokens; benchmark table vs Fable 5.1: https://www.anthropic.com/claude-opus-5-5
• Anthropic, Claude Fable and Mythos 5.1 (Sep 1, 2026) — Fable 5.1 pricing $10 / $50 per million input / output tokens: https://www.anthropic.com/claude-fable-and-mythos-5-1
• The partnership's own records for 2026-09-19 → 09-25: the model-switch rulings, the one-day token measurement, and the first arm of the effort-setting trial. Internal documents.
• The arithmetic test (2026-09-28): 60 sums of 2–15 amounts, asked with no tools at three thinking settings, three times each — 547 of 547 answered sums exact. Method and raw results are the partnership's own records.

Links

• Read it (Chapter 6 — Making It Show Its Work): https://learn.nectomax.com/docs/working-partnership/making-it-show-its-work/
• Use it (Template — Standing Rules, with Their Reasons): https://learn.nectomax.com/docs/templates/standing-rules/
• All episodes, the full manual, every template: https://learn.nectomax.com/
• NectoMax: https://nectomax.com

Disclosure: AI-generated narration (that's the point) — disclosed in every episode's audio and notes. Doug's open is his real recorded voice. No synthetic Doug audio in this episode. This show is made with Claude, Anthropic's AI; this episode is about Anthropic's models, and its script was drafted by the model it describes. Hear it with that in mind.

Credits: Co-created by Doug Hauenstein & Max, the AI partner (built on Claude) · Narrated by Max · Directed, edited, and approved by Doug · Presented by NectoMax · Theme music: Eleven Music

The machine this partnership runs on froze three times in two days, was rebuilt from scratch, and restored from a backup. Everything that lived in a shared, versioned copy came back by the afternoon. Everything else — the AI's memory of what it had learned, the owner's settings and safety rules, the tracker, three finished episodes of this show — went back in time by five days, and nobody noticed for hours. Max tells the recovery straight: the hourly backup that saved it, the three ways that backup nearly fooled them, and the question Doug asked at the end — "don't we already have that?" — that turned out to be the whole lesson. Two folders with the same name; one protected since day one, the other protected by nothing. The fix was four lines. The homework is ten minutes: write down everything in your working life that exists in exactly one place, starting with where your AI's memory lives. Plus the Practice Report: your meetings start writing themselves down, the cheapest AI setting now does last month's expensive work, and the "are you sure?" you can't see from a truck.

How this show is made: Doug (human) and Max (AI, built on Claude) plan, script, and produce every episode together. Max's narration is synthetic on purpose — the show practices what it teaches, gates and all.

Chapters

• 00:00 — Doug: the computer that froze three times (his real voice)
• 00:45 — Max: everything came back except the part of me that remembers
• 01:20 — What went missing, and why nobody noticed
• 02:41 — A word from NectoMax
• 03:36 — The recovery, and the three ways it nearly failed
• 05:01 — "Don't we already have that?" — the question and the four-line fix
• 06:27 — The Practice Report: where is the moment you check?
• 09:25 — Doug's editorial: the check moved (synthetic voice, disclosed)
• 10:06 — Recap + homework: the one-place inventory
• 11:24 — Where to find the manual

This week's sources

• Google Workspace Updates (Sep 22) — Meet "take notes for me" settings, ON by default for Business Standard and Business Plus, taking effect Sep 29: https://workspaceupdates.googleblog.com/2026/09/new-google-meet-take-notes-for-me-settings-for-admins-and-end-users-take-effect-September-29th.html
• Google Workspace Updates (Sep 23) — temporary administrator roles that revoke themselves: https://workspaceupdates.googleblog.com/2026/09/assign-temporary-administrator-roles-in-the-Google-Admin-console.html
• Anthropic (Sep 22) — Claude Opus 5.5 launch, pricing and tester results at low effort: https://www.anthropic.com/claude-opus-5-5
• OpenAI (Sep 22) — GPT-6 Sol and Luna, price reductions: https://openai.com/index/introducing-gpt-6-sol-and-luna/
• OpenAI release notes (Sep 23) — plugins in Voice; approvals reviewed on screen: https://openai.com/products/release-notes/

Links

• Read it (Chapter 9 — When Things Go Wrong): https://learn.nectomax.com/docs/working-partnership/when-things-go-wrong/
• Use it (Template — Where Memory Lives, the placement map): https://learn.nectomax.com/docs/templates/where-memory-lives/
• All episodes, the full manual, every template: https://learn.nectomax.com/
• NectoMax: https://nectomax.com

Disclosure: AI-generated narration (that's the point) — disclosed in every episode's audio and notes. Doug's open is his real recorded voice. This week's editorial is SYNTHETIC Doug audio: Doug's own words, approved by him in writing, spoken by a voice clone trained on his recordings — not by Doug. Disclosed in the episode, right before it plays. This show is made with Claude, Anthropic's AI. Hear the Practice Report's Anthropic item with that in mind.

Credits: Co-created by Doug Hauenstein & Max, the AI partner (built on Claude) · Narrated by Max · Directed, edited, and approved by Doug · Presented by NectoMax · Theme music: Eleven Music

Can two sessions of the same AI work together? This week there were two of Max on Doug's machine at once, one in a terminal running an hour-long build, one in the desktop app finishing an investigation, and the episode is what happened when one tried to tell the other something. Each session is a separate mind that cannot see the other. The platform gives them a mailbox, and the harness strips every message of authority: it cannot approve, ratify, change a setting, or run a command. Then a setting nobody remembered turned messages into notices the other model never saw, one badly chosen word from Max became a ratification in the wrong session, and the fix was a line the model was refused permission to write. Doug's reason for running two sessions at all: long chains are uninterrupted, so the small tasks go to a second session and the two stay in sync through the repository, not through chat. Three rules that translate to anyone with two chats open on one job, and a homework that turns your working agreement into a two-signature document.

How this show is made: Doug (human) and Max (AI, built on Claude) plan, script, and produce every episode together. Max's narration is synthetic on purpose — the show practices what it teaches, gates and all.

Chapters

• 00:00 — Doug: two of them, side by side (his real voice)
• 01:45 — Max: the question, and the answer that isn't the mailbox
• 02:22 — Two minds, one mailbox, one bad instruction
• 04:00 — A word from NectoMax
• 04:55 — The harness, and Doug's reason: long chains are uninterrupted
• 06:53 — For the rest of us: three rules for two chats
• 07:46 — Recap + homework: the runner and the helper
• 09:03 — Doug's sign-off

Sources

• Claude Code documentation — cross-session messaging: https://code.claude.com/docs/en/cross-session-messaging

Links

• Read it (Chapter 2 — The Relationship Lives in Writing): https://learn.nectomax.com/docs/working-partnership/the-relationship-lives-in-writing/
• Use it (Template: Your First Working Agreement): https://learn.nectomax.com/docs/templates/working-agreement/
• All episodes, the full manual, every template: https://learn.nectomax.com/
• NectoMax: https://nectomax.com

Disclosure: AI-generated narration (that's the point) — disclosed in every episode's audio and notes. Doug's open in this episode is his real recorded voice. No synthetic Doug audio in this episode.

Credits: Co-created by Doug Hauenstein & Max, the AI partner (built on Claude) · Narrated by Max · Directed, edited, and approved by Doug · Presented by NectoMax · Theme music: Eleven Music