AI Gives You Possibilities. Evidence Still Gives You Certainty.
Field notes from Boston Startup Week
I came out of Boston Startup Week with a notebook full of scribbles and one throughline connecting almost all of them: AI has made it absurdly cheap to generate — leads, decks, market theses, outreach — but nobody has made it any cheaper to be right. The teams that win aren't the ones producing the most. They're the ones who still know the difference between a signal and their own noise.
Here's what stuck with me.
Separating real market signals from AI-generated noise
Anthony Franklin framed the shift better than I'd heard it put before: software gave us repeatability, and AI gives us possibilities. But those aren't the same kind of gift. Certainty demands evidence; variation helps exploration. The problem is that AI is a variation machine, and it's easy to mistake a confident-sounding output for an actual finding.
He mapped it as a quadrant — how much value you're getting from generative variation on one axis, how much you need external evidence on the other. Depending on where you land, you're doing one of four things:
Generate — variation is valuable and you don't yet need proof. Brainstorm freely, let the model run.
Investigate — you want both breadth and rigor. Explore widely, then go verify.
Accelerate — low need for new evidence, so let AI speed up work you already understand.
Validate — you don't need more ideas, you need proof the one you have holds up.
The practical takeaway I'm actually going to use: make the model show its work. In your next session, ask the LLM to label its facts with a source, its claims with reasoning, and its manufactured answers as manufactured. It won't be perfect, but forcing that distinction is how you stop treating a plausible sentence as a fact.
A customer claim is worth nothing without behavior
This one hit close to home for anyone in sales or BD. People will endorse an idea all day and never act on it. "Would you buy this?" is one of the least useful questions you can ask, because saying yes costs nothing. What you actually want is observable behavior — what have they already done, paid for, switched away from, complained about — not a hypothetical intention.
The AI version of the same discipline: turn your thesis into an assumptions tree. The model is great at generating the tree — every belief your plan quietly depends on. But it should not be the thing that answers it. You go find the evidence for each branch yourself.
And a note on selling into skeptical rooms: if a community has a bias against, say, AI-built decks, don't try to shove it down their throats. A bias is harder to sell against than a product. You either meet people where they are or you pick a different room.
More leads won't save you
The GTM session was a gut-check. The headline stat: outbound noise has roughly quadrupled, while reps still burn three to four hours a day just prospecting and contacting. Pouring more volume into that isn't a strategy — it's a leak.
That's the funnel-versus-loop distinction. A funnel leaks: the more you put in the top, the more falls out the sides. A loop compounds — each cycle removes risk and feeds the next one. The goal isn't more leads. It's a GTM engine that gets better the more it runs.
The warm outbound playbook
The version that compounds looks less like spray-and-pray and more like this:
Define the account. Build a real ICP / early-customer profile — company criteria you can actually filter on. Tools for enriching and filling the gaps: Clay, Crunchbase, and Claude for enriching target accounts.
Define the buyer. Not just "someone at the company" — nail the seniority, function, or exact title, or get a named list.
Find the warm path in. GitHub and Slack are underrated for this — pinging your network to see who already knows someone at the target account turns a cold hit into an easy intro. Apollo for the rest.
Wire it to a CRM. None of it compounds if you can't track the campaign and integrate the loop.
And the distinction I keep coming back to: a trigger versus a signal. A signal is a state of the world — a company just raised. The trigger is what it implies: they're probably about to hire, look for office space, and start spending. The reps who win read the signal and act on the trigger before the noise catches up.
