Tolu

In the age of AI, the businesses that will survive are not wrappers, they are the ones that have done the hard graft, the messy part no one has time or patience for

You can read the history of computing through whatever metric vendors lead with. MHz when speed was scarce, cores when parallelism was, $ per token when budgets were, now performance per watt. The headline metric is a running record of whatever customers just ran out of

Muse Spark 1.1 is now live on OpenClaw!

The IBM story is really a hardware inflation story. Clients moved end-of-quarter spend into servers, storage & memory ahead of price hikes and let software deals slip. One of the first clean examples of AI costs cannibalizing other IT spend at that scale

@pmarca It will go full circle. Eventually it will all be AI talking to other AI. Then we'll create a new platform with only humans allowed. rinse & repeat

AI writing is so good now, there are only a handful of idiosyncrasies left to point out. Those will vanish shortly.

I spoke to IBM CEO Arvind Krishna about the warning today and here’s what he told me: “A few large capex deals paused near the end of the quarter. It’s not widespread. It’s mainframe and the software associated with it….Mythos is making people pause to say, wait how much do I need to spend on cyber? They’re pausing on new deals until they know…We don’t see our software being disrupted by ai at all.”

@rcwillenbrock @alliekmiller Agreed.

@LFCtolu @alliekmiller businesses that never penciled at 10 people work at 3, which means the smaller unit shows up both inside existing companies & as entirely new entities. in pe terms it changes what's even acquirable, a 3-person operation with real revenue was never a target before.

Everyone has a theory on how AI will impact the job market. Here's what I'm seeing: 1/ Everywhere I look, teams are getting smaller. That does not mean overall headcount is reducing or unemployment is increasing, but my anecdotal experience is telling me that the smallest meaningful operating unit is shrinking by a factor of 2 or more. 2/ Some government groups I have spoken with (not just in the USA) are preparing for a surge in unemployment benefit claims. That does not mean it's happening, nor does it mean their plans are sound, but it's helpful to know that is one outcome they're considering. 3/ Everyone has incentives. The AI labs, the CEOs, the researchers, the banks, the government, heads of state, even you reading this. It's important to keep that in mind, especially as tunes change (ex: from mass job loss to mass job gains). 4/ Even if AI creates new jobs, it is likely that they will require new skills or be in new areas of expertise. That means there will still be a skills gap. That skills gap feels like it's growing wider every day. More jobs existing is not the same as high employment aka it's not just about jobs existing, it's whether they get filled (and yes, I wrote that and not AI). Outside of a few hundred content creators and a couple of enterprises, I'm not seeing enough movement to close the gap. 5/ Keep pushing enterprise leaders to provide a timeline on their job predictions. They'll refuse at first because they are PR-trained. Push back. Ask for a range, ask if it's over or under 10 years, ask what would have to be true for it to happen in 3 years versus 10 versus 50.

The era where you just made the model bigger is closing. This narrative fits the broader pattern today, the models commoditize & the next round of gains comes from the engineering underneath

@garyvee Most days yes, some days I’m too tired so not really lol. And honestly that’s ok guys

I know people making $8 million a year that are fucking miserable and hate their life. and I have friends who make $67,000 a year, loving life.. four softball teams and love it. I don’t give a fuck how much money you make. Do you feel pumped as fuck when you wake up??

@northead @erikbryn @CAIS Super interesting, thanks for sharing this

@LFCtolu @erikbryn Indeed! The good @CAIS folks are working on something along those lines x.com/CAIS/status/20…

We must act now. AI capabilities are advancing far faster than our understanding of the economic implications. We must act now to guide AI to complement humans rather than simply imitate them — and to generate prosperity for the many, not just the few. I'm delighted that 16 Nobel Laureates, over 200 top economists, and top AI researchers agree and signed our statement on transformative AI. 1/n

The FINRA comparison is interesting because FINRA works off a chokepoint - you can’t touch the securities market without membership, so expulsion is a huge deal. Curious what the equivalent hook would be here? Voluntary pre-release sharing will hold up until the race gets tight, and the labs that are most worth screening are the ones with the least reason to opt in, no? If membership gated something real, compute contracts or cloud distribution maybe, this works. Without it, it’s kinda becoming peer review with extra steps?

Model commoditization is now happening pretty fast & businesses are seeing cost reductions as a result. We aren’t just seeing cost per token being competed down, but we are also seeing less tokens needed per task as the models get smarter. Microsoft’s small business Copilot now works out to ~$9.50 a seat, down from the $30 add-on. And Nvidia is out pitching your own knowledge as the moat, which is what you say when the thing you sell is becoming the electricity. For the average small business, this is kinda simple - the AI inside the software you already rent keeps getting better and cheaper without you doing anything. The part that won’t improve on its own is whether your business’s know-how is in any shape for these tools to use. That’s the only AI investment an owner actually controls

I think the overhang persists for a different reason though. Most teams aren’t short on ambition, they’re carrying product decisions made around model weaknesses that no longer exist - guardrails, chunking, human review steps that were rational two releases ago. The 3-month cadence means your architecture keeps encoding stale assumptions about what models can’t do, and unwinding those is slower than raising ambition. The overhang mostly gets forfeited to your own old scaffolding tbf

every ~3 months, you need to increase your level of ambition in the AI era, else you forfeit the capability overhang of the models to your competitors

Yeah, so basically per-token price stops meaning much once models differ this much in how many tokens they need to finish the job. The smarter model wanders less, fewer retries & shorter trajectories, so 2x per token still comes out cheaper on the bill. I think pricing pages eventually follow - $ per completed task is the honest comparison, per-token mostly survives because it’s easy to print.

Can attest. Use of voice AI is on of the areas I'm most excited about for small businesses

Part of why founders keep choosing grand & vague is who they're talking to imo. Precise and small sounds unambitious in a pitch, grand and vague sounds like a fund returner. Users select for precise, investors select for grand - and most founders meet investors before they meet users

An initial startup idea can't usually be both grand and precise. In practice they're usually either grand and vague or precise and small. Precise and small is better. You know who your initial users are, and you expand outward. With grand and vague you can't even get started.

@gxespino @andrewchen 100%

@LFCtolu @andrewchen this was my thought as well When IQ is shored up by AI, EQ will only become even more valuable

the 1-autist unicorn theory: if AI removes the need for people to build, it removes the need for people skills... this means that ultra high IQ / ultra low EQ autist incels become the most important and valuable people on the planet, going forward not just the 1-person unicorn company theory, but the 1-autist unicorn theory. QED

@arvidkahl The bigger threat probably comes from the demand side though. It's less about a competitor shipping a cleaner rebuild & more about the customer's own agents doing the job the SaaS did, one seat at a time. Code quality improving doesn't help much if the buyer stops needing seats

Overall, agentic engineering will improve overall codebase quality. People test now. They document. And while it’s easier to build a product now, existing codebases will also be hardened, optimized, and improved. So in the end, not too much of a threat for established SaaS.

So basic, but most don't internalize this yet.