The work got free. The responsibility didn’t.  ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌

September 15, 2026   |   Read online 812 lawyers and counting The work got free. The responsibility didn’t. Welcome back to the Launch Key 🚀  Full disclosure, I procrastinated this weeks newsletter. Life got in the way with a wedding, family, travel and when I finally got back to a mountain screen porch, I just wanted to want to relax on it with friends. So, yesterday I had Claude review my notes in Notion newsletter seeds, research an additional article and draft this issue in a Launch Key voice. And while Maister is something I’ve covered before, this issue is a real example of AI doing junior work for me. Eat your own dog food. Let's get into it. Read online Gmail users may wish to read online since some parts may be clipped. Take this week’s poll or comment below and let me know if we’re on the right track. Table of Contents Pull to Eject The Measurability Gap Read the job description Two caveats So how do you charge for it? The Verification Audit Old School Wisdom Modern Tools Free Knowledge Recommendations Visual Crapshoot Pull to Eject On Friday there were 2,039 of them. That’s the running count in a database kept by Damien Charlotin, a researcher who has been logging every court decision worldwide (since 2023) in which somebody filed a document containing citations an AI invented. Not alleged — found by a judge. Fabricated cases, fake quotes, authorities that do not exist, filed in more than seventy jurisdictions. Just under 1,400 of them in the United States. AI Hallucination Cases Database – Damien Charlotin The most comprehensive database of AI hallucination cases in law: legal decisions from courts worldwide, searchable by country, party, AI tool, and outcome. Updated daily.   The sanctions have been mostly modest. A few thousand dollars, a formal admonishment, a referral to the bar. The real cost isn’t the fine. It’s being the attorney who didn’t read what he signed. Notice the thing that matters the most. Every one of those filings came out of a tool that is genuinely excellent at writing legal documents. The drafting wasn’t the failure. The drafting was terrific. The checking was the failure. That gap has a name now, and there’s an economics paper behind it. The Measurability Gap In February, Christian Catalini at MIT, with Xiang Hui at Washington University and Jane Wu at UCLA, published Some Simple Economics of AGI. It’s short, it’s readable, and I haven’t been able to stop thinking about it (see Mastery). They model this transition as two cost curves moving in opposite directions. The first is the cost of doing the work. Falling exponentially, and everybody can see it. Drafting, coding, analysis, design, research — the cost of production is heading for zero. The second is the cost of checking the work. That one is stuck, because it runs on human attention, and human attention has not had a meaningful upgrade since the Pleistocene. The space between those two curves is what they call the Measurability Gap. It widens every quarter. Here’s the consequence, and it’s the whole issue: when execution becomes abundant, it stops being the constraint. The new constraint is what the authors call verification bandwidth — the capacity to validate, audit, and underwrite responsibility for work a machine produced. And when the constraint moves, the money moves with it. Value stops accruing to whoever can generate the output and starts accruing to whoever can vouch for it. That isn’t a motivational sentiment. That’s a paper about where the money goes. Read the job description Look at who the latest AI job descriptions are for: Someone who can look at a finished, confident, professional-grade deliverable and tell — quickly, without rebuilding it — whether it’s right. Someone who knows which parts of it are load-bearing and which parts are decoration. Someone who has watched this exact category of work fail before, and remembers the shape of the failure. Someone willing to put their name on it with something real at stake if it’s wrong. You cannot prompt your way into that. There is no course; the tools are the easy part (see AI tools for the non-technical entrepreneur). It is the residue of having been accountable for outcomes, repeatedly, in a specific domain, over decades — including the two or three times you were wrong in a way you still think about at 3 a.m. Somebody put it on X this summer better than I’m going to: a person with extraordinary judgment and mediocre writing can suddenly communicate at an elite level. That’s true. And the reverse is not true, which is the entire game. AI can lift the prose of someone who has something to say. It cannot supply something to say to someone who doesn’t. Two caveats There are two places where this argument is thinner than advertised. First: not all experience compounds. Deep knowledge of a specific platform, a specific framework, a specific vendor’s product — that depreciates, sometimes fast. What compounds is the layer above it: how projects fail, how buyers actually decide, which promises never survive contact with a renewal date. If your thirty years is mostly tool knowledge, your moat is narrower than you’d like, and that’s worth knowing now rather than later. Second: the authors don’t promise this ends well. They lay out two paths. In one, verification capacity scales alongside capability and you get what they call an augmented economy. In the other it doesn’t, and you get a hollow one — enormous output, nobody accountable, trust quietly draining out of everything. Nothing guarantees the good version. But both versions price verification at a premium. That part holds true either way. So how do you charge for it? You charge for it by accepting accountability that other people are quietly avoiding. Watch the language of high-value work shift in real time. Review. Audit. Second opinion. Fractional oversight. Sign-off. Those sound like the consolation prizes of the AI era. In Catalini’s framing they’re the opposite — they’re where the rents land. The phrase in the paper is insure outcomes rather than merely generate them. That’s a service you could be selling next month. No product. No platform. No code. The Verification Audit Four questions. Twenty minutes. On paper. In your old domain, what does a plausible-looking wrong answer look like? Be specific. The forecast that’s off because an assumption never got restated. The vendor contract that reads fine until renewal. The protocol that’s clinically sound and violates a state rule. If you can name three from memory, that’s your inventory — and you can, because you’ve caught them before. Who is producing that work right now, with AI, at speed, unsupervised? Somebody is. They’re faster than you, cheaper than you, and their output looks excellent. Here’s the reframe: that person is not your competition. That person is your customer. What would you put in writing? This is the real filter. Not “I have views on this.” What would you sign? Where would you accept consequences for being wrong? Whatever survives that question is a product. Everything else is a conversation. What does it cost them to be wrong? Price follows exposure, not effort. Two hours spent reviewing something where an error costs $40,000 is not a two-hour engagement. Experienced people price reviews like labor, almost every time. It isn’t labor. It’s underwriting — and underwriting is priced off the risk transferred, not the clock. Question three is the one that will keep you up. Do that one first. Charlotin’s database had 2,039 entries on Friday. It’ll have more by the time you read this. Eight hundred and twelve of them were lawyers. People with the credential, the expertise, and the tool — who skipped the part that turns out to have been the entire job. Most of the rest were representing themselves. Nobody in the room had done this before, and it showed. The work got free. The responsibility didn’t. And the responsibility is the thing you’ve been carrying for thirty years, usually without anyone bothering to call it a skill. They’re calling it one now. Now go launch something 🚀  ❝   The first principle is that you must not fool yourself — and you are the easiest person to fool.   Richard Feynman   CrowdHealth - Healthcare Crowdfunding Tired of health insurance policies that are confusing and expensive? Join CrowdHealth and finally break free from the insurance system. CrowdHealth is not health insurance. It's an innovative model to use and pay for health care without insurance middlemen and government influence. Old School Wisdom David Maister worked out the economics of this in 1993, and he had never heard of a large language model. Managing the Professional Service Firm is built on what he called leverage — the ratio of junior to senior people on an engagement. Juniors do the work. Seniors decide what work is worth doing, check it, hold the relationship, and carry the risk. Profitability comes from pushing as much execution down as possible. (see Agency of One) Every law firm, consultancy, agency and accounting practice on earth has run that model for a century. AI didn’t break the model. It broke the ratio. The junior layer is now nearly free, infinitely patient, and available at eleven at night — which means one senior person can run the leverage of an entire firm. Worth remembering that “senior” was always defined by who signs, not by who drafts. Managing The Professional Service Firm David Maister - Drawing on more than ten years of research and consulting to these unique and creative companies, David Maister explores issues ranging from marketing and business development to multinational strategies, human resources policies to profit improvement, strategic planning to effective leadership.   Modern Tools Build the reviewer, not the writer. Everyone’s first custom assistant writes things. Yours should check things. In Claude Projects or a Custom GPT — either works — set one up and paste your actual standards into the instructions. Not “be helpful.” Your standards. What you look for, what disqualifies a piece of work, the three mistakes you’ve watched people make for twenty years. Something like this, adapted to your field: You are reviewing work in [my field]. Before you say anything about quality, list every factual claim and mark each one Verified, Unverifiable, or Suspect. Then list the assumptions that were never stated out loud. Then tell me what a hostile expert would attack first. Do not praise anything. Then run the AI’s own output back through it. It’s uncomfortable the first few times. That’s how you know it’s working. Claude · ChatGPT Free Knowledge The paper behind this issue is free, on arXiv, and lighter on math than you’d fear. Some Simple Economics of AGI — Catalini, Hui and Wu, February 2026. Read the abstract and the verification section even if you skip the modeling. It’s the clearest account I’ve found of why the thing you’ve been doing on instinct for thirty years is about to have a market price. Some Simple Economics of AGI (arXiv) Recommendations 📕 MGMT Playbook : Practical management insights straight to your inbox every Wednesday. 🚏 The Cautionary : Learn to challenge conventional wisdom. 🏢 Retirepreneur : Your guide to turning decades of experience into profitable consulting, coaching, or course businesses, without risking your retirement security. 🗃️ Dealroom Business Success Uncovered : Learn directly from billionaire entrepreneurs on how to grow a business. Join a community of 2,000+ innovators. 🧾 Redefining Retirement : Too Young to Retire, Too Old To Take Orders? Yeah, Me Too. Visual Crapshoot   You've Spent 30 Years Building Expertise. It's Time to Turn It Into Income That Works for You. The complete framework for professionals 50+ who want to build multiple revenue streams without quitting their job or betting everything on a startup. Help me help you: What's the #1 challenge holding back your late career side hustle? I'm creating resources specifically for late-career entrepreneurs. Which topic would help you most? Financial risk management - can't bet retirement Worry that you're not tech savvy enough Have connections but unsure how to use them Business model selection confusion Time management with family obligations How to validate product while still employed Age discrimination concerns Is it too late for my idea? Other Share Launch Key | The Newsletter for Late-Career Entrepreneurs Thanks for referring the Launch Key newsletter! 🚀 You currently have 0 referrals, only 1 away from receiving Free tools to launch your new idea or digital product. Instant PDF download with 1 referral. Click to Share Or copy and paste this link to others: https://www.thelaunchkey.com/subscribe?ref=MRjCg0GwZI Launch Key readers – thank you for your support and feedback. I appreciate each and every one of you as I work to build something you value. Remember, if there's anything you'd like to share — a recommendation, a story idea, or just a note to say hi, hit the reply button and fire away.   ~ Rob   Update your email preferences or unsubscribe here © 2026 Launch Key | The Newsletter for Late-Career Entrepreneurs 228 Park Ave S, #29976, New York, New York 10003, United States Powered by beehiiv