Nothing makes my blood hotter than watching a real company get bent over by an AI consultant who has never shipped a single f*cking line of production code. You didn’t fail at AI. You got robbed by people who charge millions because you don’t know any better. Listen, I know this, because I've worked with the "big" three consultancy firms. Here’s the con without the embellishment. They sell you a pilot they already know dies in a conference room. They bill you to “learn your business,” which is three juniors googling your industry on your dime at partner rates. They shove the tools that pay them kickbacks, not the ones that would actually work. Their main goal is to stay, sucking the budget out of you. They slap a logo on someone else’s model, call it a proprietary platform, and dare you to open the hood. When it shits the bed they blame the model. Never the engagement. Never themselves. This isn’t a couple of scumbags. The data from 2026 shows a massive structural divide: while a vast majority of corporations are abandoning general-purpose or "assistive" AI projects due to poor returns, a highly disciplined 20% minority is capturing nearly all of the economic value. Regulators have started fining companies just for lying about what their AI can do. It's by design. And the actual work never happens. Nobody checks if the outputs are to be trusted. Nobody writes down what “good” even looks like. Nobody turns the demo into something that runs every week without the consultant still in the Slack. It rots into a zombie project, the budget is gone, and the whole room nods along to “AI didn’t work for us.” AI worked fine. The lie was the business model. We do the opposite at @LimestoneHQ. We sign an NDA and start on our own time. Not your invoice. We watch how your people actually work before we scope a single thing. We write the SOW only when we can show you what we’d build, what it costs, and what changes. Billing starts when we hit work that moves the business. Discovery is free. Then it’s an AI Velocity Pod at $17K a month, month-to-month. We don’t deliver, you don’t pay. One example: a PE-backed healthcare company had five people on a module rebuild. We put one pod on the same scope in the same codebase. One senior engineer paired with AI agents, plus a half-time delivery architect on review and quality gates. 1.5 people at $17K a month against a loaded squad burning $610K to $880K a year. Some PE partners can't believe that's our price. Ninety days later: 50% faster cycles. 122 PRs merged. 104 of them with fewer than five reviewer comments. Deployment from weeks to days. Agents wrote 98% of the code. The senior engineer reviewed every line and nothing shipped until he could prove it worked, explain why, and defend it with the agent closed. We've saved your most valuable people 15 hours per week for a fraction of the price of the so-called "consultants". That’s the difference between selling hours and shipping systems. One hides the ball. The other hands you the ball and walks out. Honest work gets you ACTUAL results, measured to the last cent.

We Must Pace the Frontier: I’ve written a new essay on why the AI industry should slow down, with a three-part plan for doing so. Anthropic is unilaterally committing to the first of these steps. We’ll provide third-party evaluators with permanent, employee-level access to our systems, so that they can verify adherence to our safety measures, report on incidents, and assess models’ alignment during training. You can read the full post here: