Love it 😍

The model is a commodity. Every day there's a new one, marginally better. Your proprietary data is the moat, said Tom Morkes at @shift_AI_events.

Helply tried three pricing models. Seats plus outcomes: two bills.
 Outcomes only: endless debate on what counts as resolved.
 Banded AI credits: debate gone. 100+ signups, about 3x revenue per account vs the old seat model. Who else moved off seats?

Groove was doing $4.5M with 1,400 customers. Tom Morkes' team closed sign-ups on purpose to rebuild around AI. The hardest part wasn't building the new thing. It was killing the old one. @Shift_AI_Events

« In Al, you have to show up with a world-class product. If you can't, it's time to pack it up. » Tom Morkes - Helply

First @shift_AI_events Europe in Barcelona. @alextheuma's opening advice: give first, help other people, and don't arrive with big expectations.

On my way to Barcelona to attend to the first edition of @Shift_AI_Events by @alextheuma.

Every time I check a GEO tool’s pricing page, Claude is locked behind the highest-tier plan. What’s the rationale behind this? Is Claude significantly more expensive to track, or is it just a pricing strategy?

I've got a big fomo right now, not to be in SF for #sftechweek

Building a company in 2028

Every time I see someone working at @AnthropicAI, it feels like they previously worked at @stripe. I’m sure there are plenty of other patterns like this across tech. Would be fun to map them all.

Back in the days, lots of people were "live tweeting" events. It was very painful to do. Wondering if someone has automated this using AI.

I’m watching more and more YouTube videos that include “real songs”. How do you manage not to be blocked? cc: @Bencera

Tried a SaaS built by a founder I really admire. It didn’t deliver for me, so I decided to leave. The dark patterns on the way out were more disappointing than the results. I wonder if that’s part of what it takes to grow. I hope not.

I’m building a new product on the side, with the best tools and frontier models I can get my hands on. Some days, it still feels impossibly hard. Oddly, that’s what keeps me going. If a single prompt could build it, a single prompt could copy it.

Could we build a “Pizza Index” for AI labs to anticipate big OpenAI, Anthropic or other labs launches? There must be public signals hiding in plain sight.

Something shifted last week in how I think about what's possible with AI infrastructure. A new surface area entirely. The builders who move early on foundational releases like this tend to define the use-cases everyone else copies later. We're still in the first chapter.

I’m starting to think the Jev hype says more about how badly we’re building AI agents than about Jev itself. We’ve been using $5+ frontier models to answer questions like: “retry?” “which tool?” “done?” “ask a human?” Of course a tiny decision model looks revolutionary. Maybe the real story is that our agent stacks have been ridiculously overengineered.

A few years ago, if you put the founder of a 1,000-person software company in a room with an early-stage founder, the learning was mostly one-way. AI is changing that. The big company knows scaling. The small team may know AI workflows, agents and product building better. That’s one of the things I discussed with @alextheuma, founder of @Shift_AI_Events, in the new SaaS Connection episode. After 10+ years building SaaStock, Alex decided to stop the brand and start again from scratch. Shift AI Europe is happening Oct 13–14 in Barcelona. 🎁 SaaS Connection listeners get 30% off with code: SaaSConnection30