this one starts with a story from fifteen years ago.
joel was in his early thirties. brazilian, new to the us, working at an enterprise — 20,000 users across five countries. his job: migrate the company from outlook and webex to google workspace.
“it took six months to convince it. then we went to legal, because there wasn’t an enterprise model to run email from the cloud yet. microsoft wasn’t pushing exchange in the cloud. webex wasn’t open to it. and here’s a guy from brazil trying to push twenty thousand people to a new email system.”
he got it done. and what he learned in that process became the spine of the whole episode.
culture and data
“what makes an enterprise an enterprise is two things. culture and data. they know how things work even without all the documents. there’s a culture in the company. it’s a substrate we need to understand and respect. and there’s data. reports and past experiences, really documented. that’s what made them survive every crisis. that’s what makes them an enterprise.”
both substrates point backward. culture is how-we-do-things-here — which is the codified history of what worked. data is what-the-numbers-say — which is the recorded history of past outcomes. an enterprise’s two main decision substrates are both made of the past.
so when a vendor walks in with something genuinely new — something that has no past inside the company yet — the enterprise’s operating system is being asked to make a forward decision using two backward inputs. that isn’t dysfunction. that’s exactly how the system was designed.
the slow buy isn’t broken
the steelman of the slow buy was the part of the episode joel was most direct about:
“if enterprises change as fast as startups change, they wouldn’t be enterprises. like startups, they would fail at 90%. you don’t see enterprises fail at 90%. they have history. they have data. their decisions are made on the past. that’s why they survive every crisis.”
the same backward-looking-ness that frustrates ai vendors is what kept the enterprise solvent through 2008, through covid, through the saas bust, through the crypto winter. you cannot get enterprise-grade reliability AND venture-grade speed from the same organism. they are coupled.
the pilot is the substrate doing its job
what an ai vendor experiences as “the eighteen-month sales cycle” is mostly the pilot. the pilot is not a stall tactic.
“the pilot is how the enterprise grows the memory of the thing it doesn’t have yet. the new entity has to enter their world. six months of pilots take six months because that’s how long it takes to produce enough rows for the dashboard to mean something. the pilot is the substrate-feeding step.”
once you see the pilot as substrate-feeding work, the time stops looking like dysfunction. it looks like the work the enterprise has to do for the new thing to actually integrate.
the closing line
rodrigo asked, late in the episode, the version of the question every ai vendor asks privately — what do you say to enterprise leaders, and what do you say to vendors?
joel’s answer at the very end was the cleanest summary of the steelman:
“if enterprises weren’t slow, there would be no vendors to sell to them. they would just build it themselves. they have the money. they have the people. the only reason there’s a vendor market is because the enterprise is what it is. so vendors complaining about the slow buy are complaining about the customer being the customer.”
what to do tomorrow morning
three asks, one per audience, which we read straight from paper for the first time on this episode:
- enterprise leader: when a vendor tells you something is new, ask “compared to what?” if they can’t ground it in something you already know, you are the prototype customer. that’s fine. price it that way.
- ai vendor: stop pitching what your product is. start pitching what it’s adjacent to. the substrate eats analogies, not announcements.
- buyer or committee: ask the vendor — “what’s the closest thing to us that has run this for at least twelve months, and what happened in month thirteen?” if they can’t name it, you have the only honest answer you need.
next
we’ve spent five weeks taking the ai industry apart. next week the camera turns around. on the operator. on the portfolio. on what running multiple companies actually looks like — not the time-management story, not the hustle story, the real one. it’s the close of the arc.
episode six: the multi-company founder thing.