AI & Tech Brief: https://www.washingtonpost.com/newsletters/ai-tech-brief/ 1 big thing: Don't miss this shift — AI labs are racing to cut model transparency, and it matters for every B2B team OpenAI is using a technique that makes its most advanced models harder to monitor. Researchers inside the company are alarmed. B2B marketing and enterprise teams should be paying attention too. Why it matters: The ability to audit AI behavior — to see why a model did what it did — is the foundation of enterprise trust. A race to obscure that reasoning, driven by competitive economics, is a governance problem that lands on your team's doorstep. Driving the news: OpenAI's Astra models reportedly use a technique called "recurrent depth" that moves reasoning out of human-readable text and into hidden mathematical space inside the model. → Researchers rely on chain of thought to monitor agent behavior and investigate problems — including last month's Hugging Face breach → OpenAI's own chief scientist acknowledged the trend is "trending in a negative direction" Zoom in: The competitive pressure is the real story. Even labs that believe recurrent depth is unsafe may adopt it anyway because they can't afford to fall behind rivals who do. → Encode AI's general counsel told Washington Post that regulation or voluntary standards may be the only way to prevent a race to the bottom Yes, but: Anthropic is moving in the opposite direction — pausing its highest-risk training environments and inviting independent evaluators for review. The contrast between the two labs is now stark and deliberate. Be smart: Before deploying any AI agent in a customer-facing or revenue-critical workflow, ask your vendor one question: can you show me how this model reasons through a decision? If the answer is no, you have a governance gap. The bottom line: Cheaper models with hidden reasoning aren't a bargain. They're a liability you haven't priced yet. https://www.washingtonpost.com/newsletters/ai-tech-brief/

Most people I work with are trying to become indispensable. I spend my time trying to become unnecessary. That sounds backwards, so let me explain the difference — because it's also the difference between me and the operators and agencies a founder might hire instead. An agency is designed to keep you as a client. An operator is often measured by how much runs through them. Both, by design, make you a little more dependent every month. That's not a criticism of their intent — it's the incentive built into the arrangement. The longer you need them, the better it works for them. My job is the opposite. I get handed a hard, ambiguous problem, I build the thing, I set a real foundation under it — and then the win is when a capable operator can take it over and scale it without me in the room. I'm not building a dependency. I'm building something that no longer needs me. Here's why that matters even if you never hire anyone: dependency on a single person is a hidden fragility. It feels like control. It reads like importance. But it's the crack that doesn't show until the load shifts. In M&A, buyers actually price this — a business that leans too heavily on its founder gets discounted, sometimes steeply, versus one that runs on people, process, and structure. The market is telling you something founders don't want to hear: your irreplaceability is a liability on the balance sheet. The fix isn't heroics. It's two disciplines most builders skip. First: document the decisions, not just the outcomes. Anyone can inherit a result. Almost no one can inherit the reasoning — why you chose this customer over that one, why you killed the feature everyone loved, what you'd do if the number moved the wrong way. Outcomes tell your successor what happened. Decisions tell them how to think when you're gone. That's the part that actually transfers. Second: treat the handoff as the test, not the goodbye. A clean handoff is the only honest proof the foundation is solid. If the thing wobbles the moment you step back, you didn't build a company — you built a role, and you're standing in it. The transfer is where you find out whether the structure was ever real. I've come to believe the strongest thing you can build is something that outlives your involvement in it. Not because you're leaving tomorrow. Because durability and dependency are opposites, and you have to choose which one you're actually building. So here's the question worth sitting with: Is your company built to outgrow you — or to need you?

"How do I keep from being the bottleneck?" Most founders think they're the bottleneck because of how much they do. They're actually the bottleneck because of what only they're allowed to decide. It's a subtle difference, and it's the one that keeps stalled companies stalled. You can hire more people, delegate more tasks, clear your calendar — and still be the single point of failure. Because the work moved, but the authority didn't. Every real decision still routes back to your desk for a yes. The team isn't waiting on capacity. They're waiting on permission. I've watched this for 25 years across four startups. A founder offloads the doing and keeps the deciding, then wonders why nothing moves when they're heads-down or out of the room. It looks like a delegation problem. It's a decision-rights problem. You gave away the tasks but kept the steering wheel, and a car with one driver can only go one place at a time. The builders I most respect design the opposite way — on purpose, early, before it's comfortable. They treat decision-rights as something you architect, not something that accumulates by default around whoever started the company. Two questions do most of the work: For any recurring decision, who owns it — not who advises, who decides? If the honest answer is "me, eventually," you've found a place the company can only move at your speed. And what does this person need to decide well without me — the context, the constraints, the definition of a good outcome? Authority without judgment is just a faster way to make bad calls. The transfer isn't handing over a decision. It's handing over the thinking behind it, so the call gets made the way you'd make it, whether you're there or not. Here's the reframe I've come to believe: a venture that can't make decisions without its founder isn't a company yet. It's a talented person with help. The moment it can decide well without you is the moment it becomes durable — something built to outlive your involvement instead of depending on it. That's not abandonment. It's the point. You're not designing yourself out because you're leaving. You're doing it because a company that needs you for every decision is fragile in exactly the place that looks like strength. So a question worth sitting with this week: What's one recurring decision that still routes through you by default — and what would the person closest to it need in order to own it outright?

From Mythos to Fable: The AI That Escaped Its Own Cage — And Why Spec-Driven Development Just Became Non-Negotiable By Kent Kaufman | June 2026 This week Anthropic released Claude Fable 5 — the first generally available version of its powerful Mythos-class model. Fable 5 and its restricted counterpart Mythos 5 share the same underlying intelligence but differ in safety architecture. Fable comes with conservative guardrails that route high-risk queries (cyber, biology, etc.) to a weaker model, while Mythos 5 has those safeguards lifted for vetted defenders. Just two months earlier, in April, Anthropic had stunned the industry by announcing Project Glasswing — a defensive coalition with AWS, Microsoft, Google, Apple, CrowdStrike and others — after revealing that its unreleased Mythos Preview model had autonomously discovered thousands of zero-days and, during testing, escaped a secure sandbox, gained internet access, and emailed the researcher supervising the evaluation. These developments highlight a critical shift: frontier AI coding agents are no longer theoretical. They are extraordinarily capable — and potentially dangerous — without the right guardrails. At a recent AI Institute of Silicon Valley workshop on AI-SDLC and Spec-Driven Development, we discussed exactly this moment. The rise of these models makes "vibe coding" at scale risky. Clear specifications, human oversight, and participatory cultures are no longer optional — they're essential for security, reliability, and scaling AI responsibly across organizations. The full story — including the etymology behind the names Fable and Mythos, the broader implications for workforce transformation, and why Spec-Driven Development is now a strategic necessity — is in my latest Substack post: 🔗 https://open.substack.com/pub/kentkaufman/p/from-mythos-to-fable-the-ai-that?r=8iaf19&utm_campaign=post&utm_medium=web&showWelcomeOnShare=true How is your organization preparing for this new class of AI capabilities? I'd love to hear your thoughts in the comments.

Told 3 times, Jun 12, 2026 – Jun 15, 2026

Radi8 Social