This week should remind all of us of the power of narratives. I’m still blown away that a substack post which was more science fiction than reality from a little known research shop called Citrini took down the whole market. And it was so powerful that well known Citadel Securities felt like it had to write a response.

No one knows the future but fear around AI is at its peak, and any story that plays into it triggers immediate “sell first, ask questions later” behavior. Claude modernizes COBOL? IBM drops 10%.

Nvidia blows out earnings? Stock drops 5%. Even with Jensen saying the agentic AI inflection point has arrived 🤯.

And this is happening even as agents are diffusing faster than we ever thought.

This FT chart captures it perfectly, three possible futures in one image: tech singularity, human extinction, or just steady trend-line growth with a small AI boost.

My bet? The blue line, but with bigger swings up and down. Short-term reactions to AI diffusion will be violent, but long-term we’ll find equilibrium.

So narratives move markets. The real question is: what do you do about it? Jack Dorsey just showed us. He admitted he overhired during COVID, the dual company structure mistake, the complexity bloat from lending, banking, and BNPL.

Then he used AI as the catalyst to cut 40% of the company – 4,000 people and in one move. Stock up 20% instantly.

This is unprecedented for a public company. And it gives air cover to every incumbent who wanted to do this but was afraid of the market reaction. Whether you believe the AI narrative or not, the market already voted.

What this shows is the time is now. You can’t wait to rip off the band-aid. Incremental doesn’t cut it any more. The world is moving too fast and in order to grow and build, you have to get lean first.

It’s all hard – hard to let people go, hard to rebuild culture, hard to come cannibalize your existing revenue with agent native products. But the alternative is dying slowly.

Shopify gets this. They’ve been infusing AI culture from the bottom up – hiring people who build with agents and letting the org transform from within (watch below).

Here’s the thing everyone gets wrong right now. In the AI era, speed wins so there’s a temptation to just keep shipping until you can’t. But sometimes the better move is to go quiet, think deeply about what’s next, invoke 7 Powers thinking, let the haters wonder what you’re up to, and then just boom 🧨.

That’s exactly what founders like Aravind at Perplexity and Howie at Airtable did — took a bit more time to do something exponential rather than just incremental.

Perplexity was left for dead by many. Then they dropped Perplexity Computer, 19 models, multiagent orchestration from one prompt, schedules jobs, thinks ahead. OpenClaw-like vibes. Multimodel will win. Must watch.

Howie from Airtable also shows what it takes. Founders driving the product, burning billions of tokens alone, then launching Hyperagent.

Guillermo from Vercel nails it – speed is table stakes now. The new edge is taste, quality, and restraint.

You can’t do any of this while bloated and burning cash. And you can’t be afraid to burn the boats and kill your existing revenue whether its AI or not.

And if you’re a smaller startup and think this doesn’t apply to you…it does. Lean teams ship faster and win. I wrote about this a year ago, and for new agent-native startups this number should be more like 75% or 100%.

This is the beginning. Narratives are powerful, perception equals reality, and whatever the truth is – every single company needs more efficiency. Now.

Control what you can. Ignore what you can’t.

As always, 🙏🏼 for reading and please share with your friends and colleagues!

Scaling Startups

#🤔

if you want the full version of the StepStone AGM for managers and for founders on how VCs and LPs are thinking, read this:

#so many 💎 in here from head of Claude Code – “Anthropic has seen a 200% increase in engineer productivity since adopting Claude Code.”

#more builders coming, time to capitalize

Enterprise Tech

#Anthropic wants to be your front door to work. The product is lights out, it has lots of integrations, an ability to build custom enterprise plugins, but just beware.

The vendor-lock in effect can be strong. And even with the ability to use one interface to access and do work across any application, does this just turn these companies eventually into a data repository, the same thing that Satya said last year when he said that every enterprise app is just a front end for a database or CRUD (create, read, update, delete) .

The question that Mark poses is the obvious one – can the traditional SaaS vendors recover from per-seat pricing or find other ways to monetize, pay for agent or data access? Great debate here 👇🏻

It’s the customer’s data but seems like the CRUD front-ends will extract their dollars.

And Anthropic is not vibe-coding SORs (systems of records) any time soon…

Akshay from Notion has a super balanced take on this which I 💯 agree with

#must read – it always starts with early adopters, the developers, but everyday workers should pay attention

“But imo, this is nowhere near “business as usual” time in software.”

#case in point

#massive news – Anthropic as supply chain risk puts it on same footing as Huawei – what are downstream implications?

#interesting that Ramp built its own security solutions after having tried Claude’s security product and all of the others from the model providers

#Wall Street sold $50B in cybersecurity stocks because Claude launched a security scanner.

Meanwhile, Claude is out here helping hackers breach entire governments because someone said ‘bug bounty.’

Maybe we still need those cybersecurity companies after all 🤔

#🎯 little bit to a torrent

#this is percentage of total calls to Anthropic API which means that mostly devs are using now and that the other industries are still lagging – will be interesting to see how fast this makeup changes over time?

#agree or disagree?

#advantage data, Google – sees 3.2x more of web to train vs. OpenAI – who wins in long run?

#Nikesh speaking his own book at Palo Alto Networks but still lots of truth to this…

#every website and application should also be built for agents first so they can easily discover, understand and use said product/service

Markets

#just be aware – this talk is rising more and more in political circles

#👀 glad the markets bounced back a little…