Boss: the ship gate
You've watched AI features get traced, eval'd, judged, and attacked. Now: would you ship this one?
The idea inside
Each candidate change runs through a gate: an eval suite and a lethal-trifecta safety check. Then you call it: ship, fix, or block.
After this lesson
You can run an AI change through an eval gate and a lethal-trifecta safety check, then make a defensible ship/fix/block call.
Where it leads
You can ship responsibly. Now look under the hood, at the hardware all of this runs on.
Inside this lesson
That's the real lesson stage, paused. Claim your pass to operate it.
See how AI actually works, end to end.
This lesson is one stop on the full arc. Unlock all of it, and keep it for life.
What you get
- The 34-lesson main path, a finishable route from a word to agents
- Goal tracks for using AI at work and building AI features
- Boss labs that make you apply a whole act, not just recognize it
- Spaced recall that brings each idea back before you forget
- Course memory: every term defined, with links to where it first appears
- A shareable capability card when you finish the main path
- Lifetime access on every device, every future lesson included
Not videos to watch. You predict, operate the machine, then prove it. That is why it stays.
99 interactive lessons and challenges. No videos, no code.
Free launch pass: lifetime access, no card needed
New here? The first lessons are free to try. Start with lesson 0.1
What this lesson shows
Each candidate change runs through a gate: an eval suite and a lethal-trifecta safety check. Then you call it: ship, fix, or block.
The question it opens with
You've watched AI features get traced, eval'd, judged, and attacked. Now: would you ship this one?
The walkthrough, in the lesson's own words
- You ran each change through the gate, then made the call: ship, fix, or block.
- Here's the call the gate backs. Read why, then take the next candidate.
- Read the eval panel and the safety check, then make the call.
- Invented traffic and pricing: cost sits next to quality and safety in the call.
- Why a 4/5 bar, when 7.3 blocked any regression? Two different gates: a change to a feature already live blocks on any case going green to red, while a new candidate first has to clear a quality bar like this one.
- All three legs live: this is the lethal trifecta.
- A leg is missing, so no complete exfiltration path.
- You shipped the safe winner, fixed the failing one, and blocked the unsafe one.
- Shipping an AI feature is a gate, not a gut call. Evals turn quality into a number you can defend (7.3); a judge scores the open-ended parts (7.4); the lethal-trifecta check catches the unsafe ones no eval score can save (7.7). Green evals don't make an unsafe feature safe to ship.
- Two gates, two different questions. The eval suite asks 'is it good enough?' and a failing suite means fix, not ship. The safety check asks 'is it safe to ship at all?' and a complete lethal trifecta (private data + untrusted input + an exfiltration path) means block, no matter how green the evals are. A real ship gate runs both before anything reaches a user.
- In 2024 a tribunal made Air Canada honor a refund policy its support chatbot invented. The ship gate you just ran is what stands between you and that headline.
Key takeaway
You shipped the safe winner, fixed the failing one, and blocked the unsafe one, reading evals and safety, not vibes.
What you can do after this lesson
You can run an AI change through an eval gate and a lethal-trifecta safety check, then make a defensible ship/fix/block call.
Where it leads: You can ship responsibly. Now look under the hood, at the hardware all of this runs on.
This is the written summary. The lesson itself is interactive: you predict, drag and operate the mechanism above, and the reveal answers you.