Crawford Del Prete

Sr. Advisor at PSG Private Equity; Former President of International Data Corporation (@IDC) - opinions are mine

[Crawford Del Prete @Craw]

@alphafox Fast. Very fast.

[Crawford Del Prete @Craw]

@FB_Helmet_Guy Jersey Generals all day long. BTW why did the Commanders not buy/license it?!

[Crawford Del Prete @Craw]

52,000 tech workers have lost jobs so far in 2026. Block just cut 40% of its workforce. Morgan Stanley slashed 2,500 roles — while posting record revenue. Everyone says the same thing: AI is taking jobs.

I think that narrative is dangerously incomplete.

A landmark Anthropic study dropped last week — the first to use actual Claude usage data mapped against 800 real occupations. The headlines ran with "75% of programming tasks are covered by AI." Here's the number nobody ran: there is currently little systematic increase in unemployment for the most AI-exposed workers. The study's own authors called for "humility."
The displacement story assumes a fixed pool of work being divided between humans and machines. But the pool has never been fixed. Not with electricity. Not with the PC. Not with the internet.

What AI is actually doing — right now, at scale — is enabling work that simply wasn't happening before. The financial model that never got built. The competitive intelligence that never got done. The analysis permanently deferred because the economics of execution never penciled out.

Clay Christensen called it "competing against non-consumption." It's the most powerful and invisible force in the AI and jobs debate — and it's getting almost zero attention.

Yes, the layoffs are real. Yes, the human cost deserves to be taken seriously. But Jack Dorsey didn't cut 40% of Block because AI automated his workforce. He cut it because Wall Street has spent three years rewarding efficiency over growth, and AI handed him the narrative.

The story of what's being created is still loading. We're making major policy and career judgments at the worst possible moment in the data cycle.

My latest piece breaks this down in full — including what the Anthropic data actually shows, why the productivity paradox is hiding non-consumption gains in plain sight, and why this time may not be as different as the headlines suggest.

Would love to hear your read on it.

https://t.co/l4KskQsRct

[Crawford Del Prete @Craw]

@stevesi Sony and others did this with magnetic peripherals. Sold hundreds…

[Crawford Del Prete @Craw]

@NostalgiaGalaxy XP IMO was the biggest jump from “unstable “- not windows native - to “pretty stable”, windows native for all. That’s what makes us look back on it with a smile!

[Crawford Del Prete @Craw]

The "AI will kill SaaS" narrative has a fundamental problem. It treats SaaS as if it were one thing. It isn't.

SaaS is a delivery and business model — not a product category. When people say AI will kill it, they're actually asking two very different questions:

- Will AI change how enterprise software is delivered and priced? Probably yes. Consumption-based models and agent-driven interfaces are already reshaping that layer.

- Will AI eliminate the need for the applications themselves? This is where the argument gets sloppy.

After nearly four decades watching technology markets get called wrong at the moment of maximum noise — the mainframe, client-server, on-premises — I've learned to treat peak disruption panic as a signal. Not to ignore the threat. But to ask a more precise question.

The survivability of any enterprise software company right now comes down to one thing: what kind of moat have they built?

Horizontal workflow tools with no proprietary data and no network effects? Harder road.

Vertical data powerhouses sitting on irreplaceable industry-specific data? AI doesn't threaten that moat — it deepens it.

Systems of record embedded in compliance and governance workflows? These don't get replaced by systems that might hallucinate.

And if SaaS is really being disrupted — where is the revenue going? We don't have a clear winner outside the existing ecosystem yet - and may not for a while.

I've written up the full framework in my latest Substack https://t.co/WC8PkdVgGc. The question isn't will AI kill SaaS. It's which SaaS companies built something worth keeping.

[Crawford Del Prete @Craw]

@stevesi Agree. But those people are (mostly) more effective

[Crawford Del Prete @Craw]

@brockpierson Yes. It was created by Digital Equipment Corporation (which is now part of HP/HPE).

[Crawford Del Prete @Craw]

Pretty straightforward travel day from Boston.

[Crawford Del Prete @Craw]

@tcrawford @mkrigsman @nyike Not to mention a governance problem!

[Crawford Del Prete @Craw]

I have noticed many of the Winter Olympic events have been modernized or added. It’s time to modernize ski jumping. I propose flying squirrel suits. Any takers?

[Crawford Del Prete @Craw]

@Gubloinvestor Hysterical

[Crawford Del Prete @Craw]

What Does a Real AI Moat Look Like in 2026?

With the rapid evolution of AI, the old playbook for defensibility is crumbling. Proprietary data, fine-tuned models, and even compute scale—once considered unassailable advantages—are being eroded by open-source innovation, synthetic data, and hardware leaps. So, what actually holds up as a durable moat in AI?

Here’s what matters now:
Data Feedback Loops & Network Effects: It’s not about having data, but about systems where usage generates better data, which in turn improves the AI, driving even more usage.

Embedded Workflow Capture: AI that becomes the backbone of business processes is far harder to displace than a dashboard widget.

Domain-Specific Reasoning: Deep, institutional expertise—baked into your models and workflows—can’t be copied overnight.

Human-AI Collaboration: The real moat is in unique protocols where people and AI learn together, building trust and judgment that competitors can’t replicate.

Integration Lock-In: The more your AI orchestrates across systems and teams, the higher the switching cost—not because of technical barriers, but because of the institutional knowledge and trust that accumulates.

Building these moats isn’t about features—it’s about compounding advantages:

Optimize for learning velocity, not just raw performance.
Invest in data labeling philosophies and exception handling.
Build trust and governance into every layer.
Focus on flywheels, not just features.

The key question for every leader:

"How much better will our AI be in 12 months than our competitors’, even if they start copying us tomorrow?"

If your answer is rooted in compounding learning, trust, and network effects—not just more data or talent—you’re on the right track.

Ready to future-proof your AI strategy?
Dive deeper and subscribe to my Substack for actionable insights and leadership frameworks: https://t.co/cNP5oIgQJx
#AI #DataMoat #MachineLearning #DigitalTransformation #Leadership
https://t.co/V4hp7utBXK

[Crawford Del Prete @Craw]

@levie IMO jobs are made up of tasks; oftentimes 100 or more . As you automate some tasks, new ones appear.

[Crawford Del Prete @Craw]

@tcrawford @sarbjeetjohal @dvellante @nyike @mkrigsman @furrier @ShellyKramer @EdLudlow @JonErlichman @jonfortt Agree @tcrawford. Raw performance is getting harder to justify; Priorities will move to things like transparency of results and explainable outcomes. This favors fewer scale providers IMO.

[Crawford Del Prete @Craw]

@ValaAfshar @Dell @MichaelDell Love this!

[Crawford Del Prete @Craw]

@dvellante @sarbjeetjohal @Cisco Agree @Cisco has a great model. In networking, it’s HW and SW welded together. Add in other SW assets and you have a great model. But there’s only so much component cost you can pass through. IMO memory is a thing for them.

[Crawford Del Prete @Craw]

@sarbjeetjohal @DellTech @HP @HPE @NetApp @Cisco @Apple @dvellante @furrier @rneelmani Yes. That’s true. But think about the sheer volume of phones, tablets and PC’s. Probably about 1.5B