One of the most important things to understand about startups is that because they grow intensely fast, they tend to induce a lot of anxiety in their founders due to perceived lack of control, and this anxiety manifests itself in various ways. A classic one is in projects that amount to ways to achieve greater legibility - hiring people to build dashboards, 'processes' to 'get their systems in order'. Almost always this is not actually addressing the real problem, which is inevitably a people problem. A perennial one is problems with sales discipline. Most salespeople are by nature crafty, ambitious, lazy, and somewhat bad with tech. They want to be on calls yapping all day and will do the bare minimum of admin that's tolerated by the business. As a result, most businesses have CRMs with terrible data quality. This results in the hiring of RevOps people, data engineers, more recently 'GTM engineers', etc. But it's fundamentally not a data problem. The problem is that the sales reps lack discipline and accountability. The founders or execs will convince themselves that if they can just get the perfect system in place with the right dashboards and fields, if they just figure out how to dedupe the data just right and build the right context layer, they'll be fine. But it's bullshit. They're avoiding the real solution, which is that they need to fire their underperforming reps and tell the rest of them that if they don't have their hygiene on point by the end of the month, they'll be fired too. In the past this was a hard ask to make because CRMs were legitimately terrible and updating them took a long time. But this is no longer the case. If you give every rep a Grok Bot, keeping the CRM updated is trivially easy. This is what actually effective AI transformation in GTM looks like, btw. Anyone who is leading with the tech is selling you a bullshit snake oil solution that will be tech debt in a few months. The real opportunity is in raising standards, driving massive amounts of urgency, and then using the new tech to find resourceful solutions to these new demands. Otherwise you'll just spin your wheels and build some pointless science projects

If 2026 was the year of vibe coding, 2027 will be the year of vibe sales, broadly speaking. The world will generally become more social, more commercial, higher energy, and more frenetic; more places in the world will come to resemble NYC or Shenzhen culturally. Your inboxes will get spammed and your phone will get called and founders will buid new companies and markets to cope with agentic outbound ramping up. Until then, make hay while the sun shines

The natural evolution of cold outbound is becoming a broker: rather than pestering someone to get on a call, you're pitching them something valuable up front. I'd expect that as AI accelerates GTM, we will move away from SDRs and towards brokers, either in-house or as middlemen

“It has always been easy to hate and destroy. To build and to cherish is much more difficult.” - Queen Elizabeth II, 1957

Think having an 'account manager' of one Bot per client may be better than one Bot per tool. In practice it's hard to segment per tool because you always end up needing multiple sources

assume it must be the case that expert network platforms have all pivoted to selling human data to labs? getting spammed with call requests every day

Surprisingly frequent way in which AI makes me more productive: - Build galaxy brain system to automate something with Grok Bot - It fails - I end up doing something manually which I'd otherwise have ignored This might seem like a failure mode but I don't think it is - ultimately the end result is still more productivity, and I'll probably end up figuring out a functional way to automate it eventually

Everyone likes the idea of thinking from first principles rather than by analogy, but it’s hard to resist pattern matching in practice. It’s rare to read an AI bear case that doesn’t rely on analogies to dot-com fiber buildout or railroad boom

“Looking at those great works of western man and remembering all that he’s achieved in philosophy, poetry, science, law making, it does seem hard to believe that European civilisation can ever vanish. And yet, you know, it has happened once. All the life giving human activities that we lump together under the word ‘civilisation’ have been obliterated once in Western Europe, when the barbarians ran over the Roman Empire.” - Sir Kenneth Clark

The data business throws off so much cash that it seems like a credible path to UBI without communism. It allows anyone to earn a decently profitable side hustle through data labelling or expert calls and makes things cheap or free that otherwise wouldn’t be (eg free house cleaning in exchange for robotics training data)

running a one-person AI services firm is incredibly fun and there's huge alpha in it

Pretty incredible to see that the guy who pulled off flying cars did in fact do it through acquiring the company that gave us 140 characters The great stagnation is officially old news. Get in loser, we're ascending the Kardashev scale

The Constellation Software Version “Pursue many small TAMs; for they stir little ambition in rivals, yet offer kingdoms to those who would rule them. Give each man his own dominion, and let a thousand captains build the empire.”

I was always sceptical of the idea of 'manage teams of agents' but I've started to come around in the past few weeks I still think you should use one agent for everything for as long as you can, but some level of specialization helpful as you scale yourself

hate to say it but the most useful consumer AI hardware would be an always-on granola recorder that surveills all my conversations and gives me follow ups from them

All my homies meeting with triumph and disaster and treating those two impostors just the same

There are a lot of incredibly smart engineers in the world who have grown amazing businesses purely through having a great product with little to no GTM. It will be cool to see what happens when they also start to develop killer enterprise products and sales teams

The interesting thing about the decentralized AI approach is that it actually favours highly manual solutions. If your reps have infinite time on their hands, then insisting on doing everything via manual data entry is ideal (ensures clear owner) Well, now they do have infinite time on their AI-assisted hands

the implications of taking seriously this idea from a 1979 IBM training manual are still not understood