Exposure Is Not Displacement

I round up the most relevant AI-in-finance news - the deals being done, who’s rolling out what, and what’s actually working on the front lines

…to embed Claude across portfolio companies. OpenAI is courting TPG, Bain Capital, Advent, and Brookfield for a $10 billion enterprise AI venture. Two labs, same thesis: PE is the distribution channel for replacing the SaaS tools it built a decade ago.

Meanwhile, Meta is planning cuts that could hit 20% of its workforce. Atlassian cut 10%. The 2026 tech layoff tally is now 45,000. And Anthropic published new research mapping what AI can actually do versus what it's being used for. The gap is enormous.

Elsewhere: Adobe's CEO stepped down after 18 years, stock 60% off its highs. Morgan Stanley says GPT 5.4 has reached human-expert level on economically valuable tasks. Gartner says IT spending will exceed $6 trillion for the first time.

But first, my take on that gap between capability and execution.

In This Week’s Issue:

From The Trenches

Exposure Is Not Displacement

Anthropic released a chart this week that should be on the wall of every investment firm's conference room. It maps AI capability against actual usage across every major occupational category. The blue area shows what AI can theoretically do. Business and finance, legal, computer science: near-maximum coverage. The red area shows what it's actually being used for.

The red area is tiny.

AI can theoretically handle 94% of tasks for computer and maths workers. In practice, it covers 33%. For office and admin: 90% theoretical, minimal actual deployment. Karpathy published a separate analysis scoring 342 US occupations. 59.9 million workers, 42% of the workforce, score 7 or higher on exposure. Jobs paying over $100K average 6.0 out of 10. The "get educated, get secure" logic inverts completely.

Anyone who's worked in high finance, PE, banking, anything related to the M&A process will look at these charts and nod. There is a lot of grunt work that AI can dramatically help with. Screening CIMs. Building comps. Normalising financials. Reviewing data rooms. It's the entire thesis behind why we built DealSage.

So the bear case writes itself. Dario Amodei warned at Davos that AI could eliminate 50% of entry-level white-collar jobs within five years. Fortune looked at Anthropic's own data and called it a "Great Recession for white-collar workers." Meta's planning 20% cuts. Atlassian cut 10%. Block cut 40%. The 2026 layoff tally is 45,000 and climbing.

If you only read the headlines, the conclusion is obvious.

But I was at the iGlobal Independent Sponsor Conference in Miami last week. During one conversation, a GP told me he'd given every analyst at his firm a Claude account. I asked the obvious question: so are you seeing the productivity gains? Hiring fewer people?

He laughed. "No. I'm actually looking to hire another analyst."

I asked if deal activity had gone up. He said no. Same volume. Same pace. He just wanted another pair of hands.

This is what I keep coming back to. The models can do 80-90% of what a junior associate does. But we are nowhere near 80-90% of associates being replaced. So what's going on? Is that a capability problem, an execution problem, or is it that replacement was never the right framing in the first place?

I think it's all three. And it's worth pulling them apart, because the answer has real implications for how you invest, operate, and hire.

The first part is execution. Anyone who's actually tried to build AI into a real workflow knows that getting a useful output from Claude or ChatGPT is the easy part. The hard part is everything that comes after. You refine the prompt. You refine it again. You handle the edge cases. You build in validation. You connect it to the right data sources. And then you discover that the workflow you've built is so specific to your context, your firm's preferences, your IC's quirks, that it doesn't transfer to anyone else. It's bespoke by nature.

That's the execution gap. It's not that the models aren't good enough. They are. It's that turning "AI can do this" into "AI is doing this reliably, at scale, across your firm" is an infrastructure problem, not a chat problem. It requires structured data, connected context, purpose-built systems. Not a login and a prompt.

But the second part is just as important: even if you close the execution gap, "replace the associates" is the wrong conclusion. And I want to be clear about this, because the lazy read of these charts is "fire 80% of your junior team." That's not what the data says. That's not how any of this works in practice. And it's not what firms should be doing.

Anthropic's own data confirms it. They found no statistically significant increase in unemployment for highly exposed workers since ChatGPT launched. The capability arrived. The displacement didn't. Not just because the bridge hasn't been built, but because exposure and replacement are fundamentally different things.

The execution gap is an engineering problem, not a fundamental limitation. It will close. Purpose-built platforms will emerge (hello DealSage). Firms will figure out how to wire AI into their actual operations, not just their chatbots. The 94% theoretical coverage for computer and maths workers won't stay at 33% observed forever.

So when the gap closes, does that mean 80-90% of associates eventually get replaced?

I still don't think so. And the reason is demand elasticity.

When ATMs rolled out in the 1970s, everyone assumed bank tellers were finished. Instead, ATMs reduced the cost of operating a branch. Banks opened 43% more branches in urban areas. Teller employment went up. The job changed, from cash handling to relationship banking, but the roles grew. The mechanism: when you make something cheaper, people don't buy less of it. They buy more.

The same logic applies to knowledge work. If AI makes deal screening 10x faster, do firms screen fewer deals? No. They screen 10x more. If diligence costs drop by 80%, do firms do less diligence? No. They diligence opportunities they would never have looked at before. I said it on a call recently: "Everybody thinks we're going to be working four-hour days twice a week. No, we're just going to be doing 10-20 times as much."

That's Jevons Paradox. The associate doesn't get replaced. The associate does 50 CIMs instead of 5. The firm doesn't shrink. The firm scales.

Anthropic's own researchers landed here too: "Even when much of a job is automated, the remaining bottleneck tasks may ultimately increase demand for complementary human skills, even among highly exposed roles."

I want to be careful not to be too optimistic. Saying "displacement isn't a concern" without specifying a timeframe is lazy. If it's true for five years but not fifteen, we've just kicked the can. Anthropic found a 14% decrease in job-finding rates for 22-to-25-year-olds entering exposed occupations. Entry-level is where the pressure shows first.

So build for both scenarios. Stop modelling AI as a headcount story. Model it as a throughput story. The value creation thesis isn't "cut 30% of staff and pocket the margin." It's "keep the team, multiply the output, grow revenue into the gap."

In diligence, reframe the question. Not "will AI replace these people?" but "what happens to this business if every employee becomes 3x more productive?" And build in optionality for being wrong. AI-contingent earnouts. Milestone-based consideration. Flexible cost structures that can adapt.

That GP at iGlobal gave his analysts Claude and nothing changed. The tool is there. The capability is proven. But without the infrastructure to turn capability into execution, the gains don't materialise. The question is how long the gap persists, and who closes it first.

News Digest

AI Labs Launch JVs to Rewire Portfolio Companies

Two stories dropped within days of each other that tell the same story from different angles. The Information reported on March 12 that Anthropic is in talks with Blackstone and Hellman & Friedman to form a Palantir-style joint venture embedding Claude AI across portfolio companies. Then on March 16, Reuters reported that OpenAI is in advanced discussions with TPG, Bain Capital, Advent, and Brookfield for a separate $10 billion enterprise AI venture. The two most valuable AI companies on earth are now competing directly for PE distribution.

The details:

Why it matters: This isn't about individual companies buying chatbot licences. This is the two most powerful AI labs structuring PE as their primary enterprise distribution channel. The race is on to become the default AI layer across private equity's portfolio.

My take: I wrote a while back that the middle market is going to get squeezed. This is more fuel to that fire. When Blackstone can embed Anthropic engineers directly into portfolio companies at scale, the resource gap between megafunds and everyone else widens further. The mid-market firm with six investment professionals and no AI budget is now competing against portcos with purpose-built AI infrastructure funded by the world's largest alternative asset manager. That said, the execution will be harder than the headlines suggest. Ripping out Salesforce or ServiceNow across a portfolio company isn't a weekend project. It requires data migration, workflow redesign, and the kind of connected data infrastructure that most portfolio companies don't have. But the signal is unmistakable: both AI labs have concluded that the fastest path to enterprise scale runs through private equity. That tells you something about where the power sits.

Anthropic-Blackstone/H&F (Reuters) | OpenAI-TPG/Bain (Reuters)

AI Layoffs Sweep Tech: Meta, Atlassian, and the 45,000 Question

The layoff drumbeat accelerated this week. Meta is reportedly planning cuts that could affect up to 20% of its 78,800-person workforce, roughly 15,800 jobs. Atlassian announced 1,600 cuts on March 11, with CEO Mike Cannon-Brookes citing the need to "self-fund AI investments." Its CTO is stepping down, replaced by two executives: one for AI and product, one for enterprise and trust. The restructuring will cost $225-236 million. Meanwhile, the running tally for 2026 tech layoffs has reached 45,000, with over 9,200 explicitly attributed to AI.

The details:

Why it matters: The pattern is no longer anecdotal. Major technology companies are systematically reallocating human capital budgets to AI infrastructure. The question is whether this stays contained to tech or spreads to every other sector.

My take: Meta isn't cutting because AI replaced 15,000 people. They're cutting because AI infrastructure costs hundreds of billions and the money has to come from somewhere. Atlassian isn't cutting because Jira writes itself now. They're restructuring to fund an AI pivot. These are capex reallocation stories dressed up as labour market stories. That distinction matters for deal professionals. When you're evaluating a portfolio company's AI strategy, ask whether they're investing in AI to replace people or to do more with the same team. The answer tells you whether management understands the opportunity or is just following the herd. The firms cutting headcount to "fund AI" without a clear throughput thesis are going to find themselves smaller and no smarter.

Read more

Other Interesting Things I’ve Read of Seen This Week:

Gartner: Global IT spending to exceed $6 trillion for first time (Mar 14) - Server spending up 36.9% year-on-year. Data centre spending up 31.7%, exceeding $650 billion. Big Tech collectively planning $650 billion in AI capex this year. The numbers are getting hard to comprehend.

Morgan Stanley warns an AI breakthrough is imminent (Mar 13) - Research note claims GPT 5.4 scores 83% on the GDPVal benchmark, at or above human expert level on economically valuable tasks. Also flags a 9-18 GW US power shortfall through 2028. Bold from a bank that's quietly cutting its own workforce.

Adobe CEO Shantanu Narayen steps down after 18 years (Mar 12) - Stock down 23% in 2026, 60%+ off 2021 highs. Beat earnings estimates and AI-first ARR tripled year-on-year. Didn't matter. When the market decides your category is cooked, even good numbers can't save you. The SaaSpocalypse's first corner office casualty.

OpenAI acquires Promptfoo (Mar 9) - AI security platform used by 25%+ of Fortune 500 companies for red-teaming LLM applications. Building the governance and compliance layer before the agentic rollout. Smart sequencing.

DiligenceSquared uses AI voice agents for M&A research (Mar 6) - YC startup doing commercial due diligence with AI voice agents plus senior human review. $5M seed round. Directly in the "AI for deal professionals" lane. One to watch.

Acquisition Intelligence is a weekly newsletter on AI in M&A for finance professionals, private equity investors, investment bankers, corp dev teams, and deal-makers.

For questions, feedback, or to share what you're seeing in the market, reply to this email.

P.S. I'm Harry, co-founder of DealSage. We're building an AI-native deal intelligence platform to help professionals turn their institutional knowledge into better decisions. If you're curious what we're up to, check out dealsage.io or just reply here