AGI timelines: what founders should actually plan for

The current timeline debate clusters in an interesting place. Serious voices put strong computer-based capability — agents that reliably do long-horizon knowledge work — somewhere in the 2026–2028 window. Physical-world AGI (robotics, real-world tasks) gets pushed further out, because atoms are harder than tokens. A loud minority insists capable-enough AGI effectively already exists; an equally loud minority points at reliability failures and says not yet.

The synthesis almost everyone quietly converges on: this is a continuous transition, not an event. There will be no morning when the headline says AGI ARRIVED. There will be a steady expansion of what a model can reliably do for an hour, a day, a week — and founders feel that as a moving line of 'jobs the model eats next.'

Whether the line crosses 'AGI' in 2027 or 2032 changes almost nothing about what you should do this quarter. The practical question isn't 'when is AGI' — it's 'what does the capability curve do to my product in the next 18 months.' That's answerable, and it doesn't require a position on the timeline debate.

The useful version: assume the cost of generating software continues to fall, the reliability of agents continues to improve on a months-timescale, and the models continue eating the thin application layer. Every serious scenario shares those three. Plan to them and the timeline argument becomes entertainment.

The same three implications keep surfacing for founders building on the ramp.

The most honest planning input isn't a timeline — it's the observation that every capability step makes distribution more valuable. When building anything costs almost nothing, the entire economy of software shifts toward whoever can get adoption. Agents writing code doesn't reduce the number of things competing for attention — it multiplies it.

So the AGI-proof plan, whatever timeline you believe: build something with a real moat under it, and build the distribution that gets it found. That plan is correct on every timeline, which is more than any AGI prediction can say.