In partnership with
What happens when you throw out the GTM playbook
That investor was wrong. Gamma is now worth $2B, with 50M users and more than half their growth driven by word of mouth.
They're one of 6 AI-native startups in HubSpot for Startups' free Bold Bets Playbook. Replit grew revenue 50x after half the team pushed back on the strategy. Ramp generated 100M+ views from a single stunt. Clay's co-founder wouldn't hang up a sales call until the prospect DMed him in Slack.
Each one took a GTM risk most founders would never greenlight. Each one paid off.
iPrompt
THE AI NEWSLETTER THAT TURNS NEWS INTO ACTION
THE HOOK
The most capable AI model ever released is sitting in your paid Claude plan right now — included until 22 June, then the meter starts. The company that built it asked rival labs for a brake pedal five days before shipping it. Your job this week is simple: find out whether it’s worth paying for. The frontier didn’t slow down. It split in two.
AI NEWS ROUNDUP
This week in AI
1 Anthropic released its most powerful model — through two doors. Claude Fable 5 is the public door: anyone on a paid plan has it until 22 June. Mythos 5 is the same model with some safeguards lifted, reserved for vetted cyberdefenders. Ask Fable something sensitive — security, biology, chemistry — and an older model, Opus 4.8, answers instead, with a notice that it did. Anthropic says that happens in under 5% of sessions. Anthropic →
2 Five days earlier, the same company asked for a brake pedal. In “When AI Builds Itself”, Anthropic revealed Claude now writes over 80% of the code merged into its own codebase, and called for a coordinated way to slow frontier development. Then it shipped Fable 5. Conscience or marketing — the internal numbers are striking either way. Scientific American →
3 Apple rebuilt Siri on Google’s brain. At WWDC, Siri became “Siri AI” — a full chatbot with its own app, running on Google’s Gemini under a deal reported at roughly $1bn a year. The ChatGPT handoff is gone, the rumoured option to choose Claude never showed, and EU iPhones miss the launch. You’ll talk to Apple; Google will answer. MacRumors →
4 Europe shows where this is going. The EU’s own cyber agency just received Mythos — the version with safeguards lifted — and OpenAI is giving vetted European defenders a freer version of GPT-5.5. The pattern is the point: verified institutions get full capability — everyone else gets the guarded edition. Tech Times →
OUR ANGLE
🔭 Capability stopped being the product this week. Access is. The fact: one model, two products. Fable 5 and Mythos 5 are the same system — Anthropic’s words, not mine — and what separates them is vetting, not capability. OpenAI says the same about its cyber model: it isn’t meaningfully more capable than standard GPT-5.5, it’s simply allowed to do more, for people who’ve proven who they are. The capability gap between the tiers is close to zero. The access gap is everything. What I infer: when models cross a danger threshold, labs don’t stop shipping — they ship trust tiers. And notice the small print: Fable 5 is “included” in paid plans only until 22 June, after which it moves to usage credits — last week’s metering thesis arriving on schedule, dressed as a gift. Even the 30-day data retention Anthropic now requires on this tier — overriding zero-retention deals — is the same trade: capability in exchange for being watched. My bet — against the herd: everyone’s reading this as a one-off safety story. I think it’s the new shipping model. By the end of Q1 2027, every major lab will sell its flagship in two tiers — a guarded public edition and a verified full-capability edition — and identity verification will become a paid enterprise product in its own right. Who you can prove you are, not what you can pay, becomes the thing that unlocks the frontier. If you think the split stays a cybersecurity niche, reply and tell me why. |
THE THREE SPECIALS
Do · Use · Understand
🎯 PROMPT OF THE WEEK
The Frontier Stress Test
This is the one action of the week. The new tier’s whole claim is that its lead grows with task length — so don’t test it on chat. This prompt packages your hardest recurring job as one long-horizon brief, then makes the model audit its own run. You leave with a budget decision, not a demo — and a one-line reply for me: what it did that your current model couldn’t, or where it fell over.
You are taking over the hardest recurring task I do. THE TASK: [describe one multi-step task that normally takes you 2+ hours — full context, inputs, and what “done” looks like] Rules: - Work it end to end. Don’t stop to ask unless you are genuinely blocked. - Show your plan in 5 lines first, then execute. - Flag every assumption you make in [brackets]. When you finish, audit yourself: - Which step were you least confident on, and how would I verify it? - What would you need from me to run this monthly without supervision? - Which parts could a cheaper model have done just as well? |
Why it works: a two-hour task is the only honest test of whether this tier changes anything for you — quick questions won’t show the gap. The self-audit at the end converts the run into the decision that matters: keep the frontier model when the credits start, or route this workflow somewhere cheaper. Where to be careful: this tier carries mandatory 30-day retention — even where enterprises had zero-retention agreements. Don’t feed it anything you couldn’t defend storing for a month. Tested on: Fable 5 against Opus 4.8 on two of our own production workflows — an informal test, not a benchmark, but the gap showed on the multi-hour task and vanished on the quick ones. |
🛠️ TOOL OF THE WEEK Artificial Analysis Launch-week claims, independently measured — intelligence, speed and price for every major model on one page. ★★★★½ / 5 Use if: a vendor just told you its new model is “state-of-the-art on nearly all benchmarks” and you’d like a second opinion before moving your workflows. Skip if: you use one chat app, on one plan, and genuinely don’t care which model answers. Every model this week arrived wrapped in its maker’s own benchmarks. Artificial Analysis re-runs the measurements independently and plots quality against price and speed — one chart tells you whether a frontier model is worth five times the cost of the one below it. For most workflows it isn’t, and this is the fastest way to find your exceptions. Describe it to a colleague: “It’s the consumer-reports lab for AI models — vendor claims in, independent numbers out.” Best use case: before 22 June, check where Fable 5 actually lands against the model you pay for today — then decide if the credits will be worth it. Artificial Analysis → |
💡 TIP OF THE WEEK Know which model actually answered you Silent routing became the norm this week: Fable hands sensitive questions to Opus, Siri answers with Gemini, Copilot picks from a menu. Quality, cost and privacy vary by model, not by product — the brand on the box no longer tells you which brain produced the answer. Three habits: Check the picker before you judge. A “worse” tool is often just a cheaper default model. Look at what’s selected before you blame the product. Watch for the fallback notice. Fable 5 tells you when Opus answered instead. Re-read that output knowing a different model wrote it. Pin versions in anything automated. In API work, name the exact model string, never “latest” — silent upgrades change behaviour mid-workflow. Why it works: knowing who answered turns “this tool got worse” — unfalsifiable, useless — into “this model underperforms on this task”, which you can act on. Where it doesn’t apply: when the router is the point — if a budget model quietly handles your routine work well, that’s a feature. Notice it, then exploit it. |
YOUR MOVE
One action. Reply by Friday.
This week in three lines:
Anthropic’s most capable model ships as two products — public and vetted — split by trust, not capability.
The brake-pedal plea and the launch came from the same company, five days apart.
Every answer now carries a hidden variable: which model produced it, what it cost, and what got stored.
Your one action: run the Frontier Stress Test on your hardest recurring task before 22 June, while Fable 5 is still included in your plan. Then reply with one line: the thing it did that your current model couldn’t — or the moment it fell over. I read every response.
Optional, after you’ve replied: the deep dive on the trusted-access era. The reply comes first.
—
R. Lauritsen
EDITOR · IPROMPT
P.S. Hit reply first — then, if you want the longer argument, the deep dive makes the case that identity verification becomes the next thing labs sell. I’d rather hear how your stress test went.
Forward iPrompt → For the colleague who still thinks all AI models are the same. |
The IT strategy every team needs for 2026
2026 will redefine IT as a strategic driver of global growth. Automation, AI-driven support, unified platforms, and zero-trust security are becoming standard, especially for distributed teams. This toolkit helps IT and HR leaders assess readiness, define goals, and build a scalable, audit-ready IT strategy for the year ahead. Learn what’s changing and how to prepare.