[ THE CURVE ]
Why "twilight" is the right word for this AI moment
We keep reaching for the wrong word. Revolution. Disruption. The cliff. Each one says the same thing: everything changes at once, and you'd better have already changed. But that's not what this moment feels like from inside a real organization. Inside a real organization, the old operating model isn't dead. And the new one isn't quite born. Both are running at the same time, in the same building, and nobody can quite see the edges.
That's not a revolution. That's twilight — the half-light between two days. And the dangerous thing about twilight isn't darkness. It's that you can still see well enough to feel confident...just not well enough to be right.
I'm calling this moment the AI Twilight, because naming the moment correctly changes how we show up in it. A revolution tells you to pick the winner fast. Twilight tells you the opposite: the winning tool will change twice before your procurement cycle closes, so betting the org on a tool is betting on the weather. The work of leadership in twilight isn't picking. It's closing the gap between how ready you feel and how ready you are...before that gap closes on you.
The antithesis this frame pushes against? The comfortable story most leadership decks tell: AI is a capability problem. Get the best model, train the staff, the value follows. In twilight that story gets you caught. Because the thing that fails first isn't the model. It's the half-light judgment. The demo that looked like readiness, the pilot that looked like adoption, the adoption that looked like trust.
In twilight, the danger isn't the dark. It's the confidence.
So the frame separates three things organizations habitually collapse into one.
Performance is not adoption. Adoption is not institutional trust.
A model can perform brilliantly in a demo, get adopted widely by the eager, and still earn zero durable trust — because trust isn't a feeling that accrues, it's a structure you decide to build. Governance. Escalation. Incentives. Who owns the outcome.
There are two trajectories out of any twilight, and you're already on one of them. One organization spends the half-light buying tools and touting their progress. The other spends it building the structure that makes any tool safe to trust. It finds, when the new day arrives, that it can move faster because it decided who holds the wheel while the light was still low.
The cliff says the change is coming. The curve says it's already here, in the half-light, and the only question is whether you're designing for it or squinting at it.
──────────────────────────────
[ THE SIGNALS ]
→ Half-Baked. [ FIRST WAVE ] The governance reflex is about to invert. Gartner's late-May read is that applying uniform governance across AI agents will itself cause enterprise AI to fail — and separately predicts enterprises will start demoting autonomous agents after production incidents expose gaps they didn't see in the demo. My half-baked extension: "do we trust agents" is the wrong frame. The real design question is graduated trust — different autonomy, different controls, per agent. That's a Twilight move: build the structure in the dim, not after the incident. Gartner, May 2026.
→ Hot Take. "Are you using AI?" is a daylight question for a twilight problem. Adoption is the easy half to measure, so it's the half everyone reports. The hard half — has the organization actually decided what these systems are allowed to do, and whose name is on the outcome — goes unmeasured precisely because it's hard. A number you can put on a slide will always beat a structure you have to argue for. That's not a measurement problem. It's a courage problem.
→ Confession. I had the value gap diagnosed backwards for a while. Two 2025 reads point at the same shadow: BCG's Widening AI Value Gap found only about 5% of companies are capturing real value from AI — roughly 60% see little to none — while MIT's State of AI in Business found 95% of generative-AI pilots delivered no measurable P&L impact, even as nearly everyone is "experimenting." I used to read that as a maturity gap: give it time, the value lands. I don't anymore. It's a twilight gap — felt readiness racing ahead of structural readiness. Time alone doesn't close it. Only the design decision does. BCG, 2025 · MIT, 2025
──────────────────────────────
[ THE NEXUS ]
Where in my organization does AI look most ready — and is that the place I've actually decided the least about who holds the wheel?
Demos go well when confidence outruns design. In twilight, your brightest-looking corner is the one most worth checking for a structure underneath the shine.
Reply if you see this differently. I read every response.
──────────────────────────────
[ THE MONDAY MOVE ]
Monday morning, run the twilight test on one system you already trust.
Pick the AI use case already humming in your org. The one nobody worries about, the one that demos clean. Now write three lines about it on a single card: what it's performing, what it's adopted for, and what you've actually decided and governed about it. Three different answers. If the third line is the shortest — if performance and adoption are full and the governance line is a shrug — you've found your twilight gap. That bright corner isn't your safest. It's your least-designed.
You don't fix it by adding a tool. You fix it by making one structural decision: name who owns the outcome when that system acts, and write down the one thing it must never do. That's the whole move. One bright corner. One card. One owner. Monday.
──────────────────────────────
[ THE COMPOUNDING ASSET ]
This week I named the lens behind everything above and added it to the canon — the AI Twilight. The page goes a layer deeper, to the three forces underneath it: why your map, your dashboard, and your headlights fail at once. → Read it.
──────────────────────────────
The grounding
This newsletter is called Signals from the Curve because there are two kinds of forecasting: the curve and the cliff. The cliff says everything changes at once. The curve says it's already changing, in the half-light, and you just have to know where to look.
I write here every Wednesday morning about what I'm seeing in AI strategy, leadership, and the parts of work that compound. Next week I'll name the discipline underneath the AI Twilight — what I call Organizational Intelligence Design — and why most organizations don't have an AI problem so much as a design problem.
If something here changed how you're thinking, hit reply. I read every response.
— Chris
Chief of AI & Strategy at Essential Innovations · Founder, Attainable AI · Taught AI Strategy at Columbia University