Bits and Bobs 3/30/2026

I found myself coming back to McKeown's Essentialism. The failure mode is so persistent. The instinct under pressure is always to do more. More bets, more threads, more motion. It feels productive. It isn't.

The conditioning runs deep. I've been trained my whole life to equate busyness with value. Longer list = more ambitious person. The insidious part is that the conditioning disguises itself as ambition. So I don't even notice when I'm adding instead of choosing.

My coach gave me a simple assignment this week: clean out half my closet. The resistance is the point. It's not about the clothes. It's about the fear of letting go when your instinct says keep everything because you might need it. That's the same instinct that drives scope sprawl at a company. Same instinct, different closet.

Here's the paradox that still trips me up: to do less, you have to explore more. You can't know what's essential without surveying the field. But exploration has a way of never converging. Everything you touch becomes a commitment. The discipline is exploring with the intent to discard most of what you find. Exploration as filter, not funnel.

Fewer bets, more energy behind each one. Close decisions, sprint, finish. The paradox is that constraint is what makes big things possible.

On the consumer side, distribution incumbents are swallowing standalone AI apps. Meta acqui-hired four AI teams in four months.

On the enterprise side, OpenAI just declared "code red." They spent 2025 trying to be everything: chatbot, video generator, browser, device company. It didn't work. Now they're pulling back to focus on coding tools and enterprise, which is where Anthropic has been the whole time. Anthropic wins 70% of new enterprise deals head-to-head against OpenAI, per Ramp's March data.

The split: consumer AI goes to whoever already has distribution. Enterprise AI goes to whoever has developer trust. The question is what happens to companies in neither lane.

Meta just acqui-hired both Gizmo (prompt-driven mini apps) and Dreamer (personal AI agents platform). Different products, same bet: software should be generated for you and shaped by you on top of a data substrate. Not apps in the App Store sense. More like disposable, contextual software that serves a need and may not outlast it. Software as utterance, not artifact (in line with Karpathy's Software 3.0 thesis)

The further you push this, the more it depends on knowing who "you" are. Without persistent context, personal software resets to generic every session. With it, "make me a planning tool" becomes "make me a planning tool that knows how I work."

Meta wants to own the whole stack: context (social graph), generation (Dreamer, Gizmo), distribution (Instagram, WhatsApp, glasses). The open question is whether the context layer should live inside one platform or be more liquid, user-controlled and portable.

If consumer gets absorbed by incumbents and enterprise consolidates around model companies, the independents that survive are the ones building both a consumer surface and an infrastructure layer.

This connects to last week's note on vertical AI compressing the B2C-to-infrastructure arc. Build the consumer habit. Expose the intelligence as APIs. If distribution concentrates, you're the vertical brain that gets called. If it fragments, you own the daily habit directly.

The test: is the intelligence core compounding across both surfaces, or are you running two products under one roof? The former is a control plane. The latter is overhead.