Uncle J's Insider
Reading paths
Human in the Loop
Division of labor, judgment, organization interfaces, and governance boundaries in AI adoption.
12 posts
Agentic Engineering
Specs, skills, agent teams, and organization-level output.
16 posts
AI × Organization
Moving management, workflows, and boundaries into AI-native orgs.
19 posts
Knowledge / GEO
Content that is readable by people, indexable by search, and reusable by agents.
8 posts
Products & Ship Notes
One-person products, launch stories, and working systems.
9 posts
AI Is Not a New Tool. It Is a New Division of Labor.
Human in the Loop, Part 1. Once the agents are running, the real question is not which model you bought. It is who decides, who signs, and who is on the hook.
Why the Org OS Cannot Run AI
Human in the Loop, Part 2. The model is not the bottleneck. The roles, the workflows, the knowledge, the responsibility lines, and the three ledgers underneath them — those are what cannot run the new division of labor.
Why the HR Three-Pillar Model Breaks
Human in the Loop, Part 3. The three pillars are not getting kicked over by AI. The old interface stopped carrying current. What HRBP actually sells is judgment translation — and that half is the first to get discounted.
The New Labor Contract
Human in the Loop, Part 4. Once Skills, Agents, and anti-distill tools show up, the contract between company and employee stops being about hours. It gets repriced around injection, judgment residue, and accountability.
AI Replaces Actions, Not Organizations
Human in the Loop, Part 5. Wulf's six interaction modes only tell you where humans and AI stand in a workflow. Cross them with five layers of organization design — role, process, knowledge, accountability, governance — and you get a 6×5 matrix executives can actually run a meeting on.
The Judgment Premium
Human in the Loop, Part 6. The cheaper AI makes actions, the cheaper actions get; the cheaper actions get, the more expensive judgment becomes. HITL is not a confirm button. It is three rights: judgment, accountability, and the final veto.
FDE Is Not an Engineer. It Is the Organization Interface.
Human in the Loop, Part 7. In May 2026, FIS, OpenAI, and Google Cloud all bet on forward-deployed engineers. FDE is not engineer-plus-sales. It is the physical embodiment of the organization protocol layer — the interface that moves judgment between the customer site, the production system, and the chain of accountability.
A/O/G: Three Cuts Into One
Human in the Loop, Part 8. The CEO does not start by asking which AI to buy. The CEO starts by deciding which knife to pick up first — the action cut, the organization cut, or the governance cut. A/O/G is not a maturity model. It is a surgical table.
The Organization OS Architect
Human in the Loop, Part 9. By the ninth essay the question is no longer about AI. It is whether the company has anyone who can redesign the organization OS itself.
Human in the Loop
A nine-essay series for founders, CEOs, and business owners. AI is not a new tool. It is a new division of labor. Human-in-the-loop is not a confirm button. It is who decides, who signs, and who can stop the system before it touches the real world.
The 11 Failure Modes of AI Agent Systems
AI agent failure modes I've cataloged after eighteen months running multi-Agent systems in production. Eleven distinct ways agent systems fail — every one of them is a management problem I saw in my HR career first.
95% of People Are Stuck Optimizing Prompts. They Are Wrong.
95% of companies are managing AI the way you manage software. Agents aren't tools — they're coworkers. The scarcest skill in the AI era isn't engineering. It's people who understand both organizations and Agents.
From HR to AI: What 18 Years of Managing People Taught Me About Managing Agents
I spent 18 years in HR — layoffs at AB InBev, building people infrastructure at SnowPlus, leading 150 at Longfor. Then I left and started writing code. Four days, 52 spec documents, 105,000 lines. People asked what made me think I could do this. They were asking the wrong question. Managing fifty AI Agents and managing fifty people are the same job.
How I Manage Agent Teams Like a Company
Last quarter I fired three AI Agents. Here's how I manage Agent Teams like an HR director runs a company: org chart, performance reviews, firing policy.
He Attached the Same 4 Pages to Every Letter for 23 Years
Bezos wrote 24 shareholder letters over 23 years. At the end of every single one, he attached a copy of his 1997 letter — the one from the year Amazon went public. Not ritual. Not nostalgia. A contract he signed with his future self, and re-signed annually in public. This is what those letters actually say.
Most "AI Agents" Are Just Chat Boxes in a Costume
Everyone is building "AI agents." Almost none of them are. After studying the Claude Code source leak, shipping my own agent SDK, and watching a founder show me a chat box and call it an autonomous system — here's what actually separates an agent from an LLM wrapper. Seven decisions. Seven rungs. One framework you can use in the next five minutes.
Wishing, PUA-ing, or Building Walls? Three Stages of Actually Using AI
I spent 18 months going from wishing AI would work to yelling at it to building systems where it cannot fail. Here is what I learned — and why 11,600 developers are still stuck at stage two.
17 Red Lines for a $2 App
Most apps ship features. One Page ships constraints. 17 red lines, zero third-party dependencies, one gesture, 30 seconds. Here's what happens when you design a product by deciding what it will never do.
The First Day I Wrote My Thesis, I Didn't Write a Single Word
Because a thesis is a system, not a document. A 320-line CLAUDE.md constitution, 12-item SSOT data calibration, 4 Phase Gate quality checks, a 1,262-line assembly script, 61 references fully verified, advisor feedback ticketized. The full systems engineering breakdown of a Peking University MBA thesis.
My Obsidian Is Not a Note-Taking App — It's an Operating System
506 git commits. 85+ AI Skills. 3 lifecycle Hooks. 435 auto-generated session logs. 7 platforms. One person. Under 60 days. How an HR guy turned Obsidian into an operating system. 14,000 words. The entire system, taken apart.
Organization Design Is Agentic Engineering
I can't write a single line of code. 4 days. 52 modules. 2,716 tests passing. 105,000 lines of code. How? I spent 18 years learning to design systems that run without me watching. That system used to be called an organization. Now it's called an Agent team.
I Quit a Perfect Partnership to Build a One-Person AI Company
An 18-year organizational operator walks out of a healthy partnership to build a one-person AI company. The math that made it inevitable.
From a Meeting Recording to a Running Project — In One Afternoon
I went from a 90-minute meeting to a fully delivered project in 4 days — 52 dev specs, 2,716 tests passing, 105,000 lines of code, and a 24KB project constitution. No team. No Jira. Just one person, one terminal, and a swarm of AI agents. This is what agentic engineering actually looks like.
From Beer to Bots — The Twenty-Year Path to a One-Person Company
In 2013 I was organizing sales rallies for 240 beer salespeople. Last week I submitted an iOS app at 11pm with three AI agents running in the background. Between the banquet hall and the apartment, there are twenty years. Nothing was wasted. But nothing was planned either.
How I Use Claude Code — Not as an Engineer, but as a Commander
Most people use Claude Code to write code. I use it to run a one-person company — three AI agents, an Obsidian vault, three iOS apps in 30 days, and 53 MBA courses distilled into 173 reusable decision tools. The tool is the same. The scope of ambition is the variable.
How Three AIs Share One Brain
Three AI agents, three platforms, one shared Obsidian vault synced through Git. The personality isn't in any model's weights — it's in a markdown file that all three can read. No database, no API middleware. Just folders, git commits, and a human gateway that approves every crossing.
AI Read 53 Courses for Me
Spent 400K+ RMB on a Peking University Guanghua MBA, two years later couldn't remember a thing. Used Claude Code Teams to command 7 AI Agents, distilling 53 courses in 72 hours — 173 Concepts, 58 Skills, 565 notes. Scaffolding is meant for you to climb up, then tear it down.
What YC Won't Tell You: 7 Signals Hidden in Their 2026 Spring RFS
RFS is not a menu. It's a signal. Don't look at what YC is asking you to build — look at what YC is revealing about itself. From 25 directions down to 7, from pure AI to AI + legacy industries, from SaaS to selling outcomes — YC is also figuring it out as they go.
The Fox and the Hedgehog in the Age of AI
For half a century, management theory said the hedgehog wins. But the environment changed. AI gave foxes the means of production for the first time. Curiosity is no longer a burden — it's leverage. This isn't a career change. It's sovereignty reclamation.
Wake Up: I Wrote My Department in Code
AI isn't an employee — it's an organizational capability module. Years of HR experience studying org design, now applied to building a controllable, auditable, evolvable virtual department with AI. The stronger the capability, the more it needs a constitution.
Subscribe to Uncle J's Insider
Notes on AI organization, agentic engineering, and content systems when they are worth sending.