If you're a technical leader, these are the questions I ask myself daily — and the ones I think more engineering orgs should be asking.

On the business itself

• When frontier models can do what your company does, what will you do? If you process payroll, two model generations from now an agent handles 100% of it end-to-end. Are you already pivoting toward what only exists once that's true — UBI administration, compliance arbitration, something else? Or are you optimizing a business with a visible expiration date?
• Which parts of your product are a moat, and which parts are just "we got here first before AI did"?
• If a competitor launched tomorrow as an AI-native version of you, with 8 engineers instead of 800, what would they cut first?
• Is your product ready for the agentic web? Humans will soon start interacting with the web via their AI tools only.

On the infrastructure you're building

• Is the infra you're shipping this quarter useful for the models of today, or for the models of 6 months from now? Building for today's context windows, today's tool-calling limits, today's hallucination rates is building technical debt with a known expiration date.
• If you're investing in MCPs and tool layers — are those tools designed for an agent that needs hand-holding, or for one that doesn't? Most MCP tooling I see today is scaffolding that future models will route around. Cluttered tool registries actively hurt stronger models.
• Where are your evaluation harnesses for agent behavior? Not unit tests. Eval suites that score whether an agent made the right judgment call. Without these, you can't tell whether a model upgrade helped or hurt.
• Is your data model legible to an agent that has never seen your company before? If onboarding a new agent takes the same effort as onboarding a new engineer, your context layer is the bottleneck — not the model.

On how your team actually works with AI

• Are your engineers using AI to autocomplete, or to generate, evaluate, and ship entire solutions? There is an order-of-magnitude difference between the two.
• Are you capturing all the feedback the model can give you? Every prompt in my workflow ends with "summarize what you changed, what you'd improve, and what I might be missing." That single addition surfaces more architectural insight than most code reviews. Standardizing prompts like this across an org is nearly free and consistently underused.
• When AI accelerates an engineer's output 5x, where does the saved time go? If the answer is "waiting for review, waiting for QA, waiting for the next sprint" — the bottleneck moved, and nobody redesigned around it. The speed gain gets absorbed by process instead of captured by the business.
• Who on your team has actually restructured their workflow around AI, versus bolted AI onto the old one? These look identical on a resume and produce wildly different output.
• Are your managers and processes designed for a world where code took days to write? Standups, sprint planning, ticket grooming, handoffs between PM → eng → QA → DevOps — most of this exists because writing code was slow. It isn't anymore.

These are the questions driving how I want to work next. If they're the ones you're already asking, or willing to ask, I'd like to talk.