Hard Questions
Why now? Can't this wait until the category is more proven?
The inertia default
You're asking this because waiting has worked for you before. Most of the technology that has tried to march through your inbox in the last decade was either a rebrand of something older or a feature that ended up bundled into a tool you already paid for. Sitting still and letting the noise sort itself was the right move more often than it was the wrong one. The inertia default isn't laziness — it's pattern-matching against a long run of times when patience was rewarded.
I want to credit that instinct before I argue against it. Plenty of categories deserve a wait. The ones that don't survive the wait usually share a tell: while you watch from the sidelines, a competitor a few miles away is accumulating something you can't buy back later. That's the part the inertia default tends to miss, because the cost of waiting is invisible until it isn't.
The slower thinking
The thing that compounds while you wait isn't the technology — that part actually does keep getting cheaper and better. The thing that compounds is the operational knowledge of how a specific agent fits a specific stack. Six months in, you know which triggers are noise in your environment, which escalations the team actually wants, which drafts get used and which get ignored. None of that knowledge is portable. Your competitor who started six months ago has six months of it. You can buy the same model they use; you can't buy the six months.
The category framing — "unproven" — also deserves a second look. The earliest cohorts of agentic deployments inside operating businesses are now well past the year mark in public production. The people running them have written about what worked, what regressed, what they had to rebuild. That doesn't make the category mature; it does mean the question has shifted from "does this work at all" to "does this work for a business shaped like mine." Those are different questions, and the second one only has an answer if you run the experiment.
The asymmetry is what tips it for me. The downside of starting now, on a tier with cancel-anytime month-to-month terms and a price that's a small fraction of the comparable loaded salary, is one quarter of a workflow experiment you decide to walk away from. The downside of starting eighteen months from now, after a competitor has compounded operational learning you can't shortcut, is a catch-up gap measured in fiscal years. The two downsides are not the same shape. See /pricing for the math; the point here is that the experiment is reversible and the wait, in some markets, isn't.
Sources
J. Dorsey, R. Botha, From Hierarchy to Intelligence, Block, 2026
What would have to be true for the opposite to be correct
- You don't have a clearly scoped piece of recurring work an agent could own — no inbox, no monitor, no recurring brief, no routine draft.
- There's no one on your team willing to be the human owner of the deployment for the first ninety days.
- Your stack lacks the integration surface (Slack, email, a documented data source) the agent would need to do useful work.
- Your competitive set genuinely doesn't compound operational learning — the work is one-off enough that a six-month head start isn't a head start.
- Your near-term capital plan can't tolerate a small recurring line item, even one that's reversible inside a quarter.
Where to next