Software Isn't Dead

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

…as both public and private investors scrambled to audit portfolios amid heightened fears of AI disruption risk.

Meanwhile, Goldman Sachs quietly announced a partnership with Anthropic to embed AI engineers directly inside the bank. Further proof that in enterprise AI, it's the partnership on deployment that matters just as much as the technology itself.

Elsewhere this week: the WSJ published a damning report on Microsoft Copilot, JPMorgan's family office survey revealed 65% want AI exposure, and Goldman led a $75M round into an AI-native audit platform the same week it announced the Anthropic deal.

But first, my take on the SaaSpocalypse and what it means for finance specifically.

In This Week’s Issue:

From The Trenches

The SaaSpocalypse and What It Means for Finance

Apollo's John Zito put it bluntly at a Toronto gathering last fall: "The real risk is, is software dead?" Lightspeed's Isaac Kim, ex-Elliott, went further: "Technology private equity, in its current form, is dead."

Both private and public portfolios have been hammered this week. Names familiar to those in finance - Thomson Reuters fell 16% in a single day. Its biggest drop on record. RELX fell 14%. FactSet dropped 10%. Salesforce is down 26% year to date. The S&P North American Software Index posted its worst monthly decline since October 2008. Down 15% in January alone. The iShares Software ETF is off 20% year to date.

The catalyst was Anthropic's Claude Cowork plugins, released January 30, automating tasks across legal, sales, marketing, and data analytics. As Schroders' Jonathan McMullan put it: "The historical 'visibility premium' erodes. The speed of AI advancement makes long-term valuations harder to defend."

The PE exposure is where this gets serious. Between 2015 and 2025, PE firms acquired over 1,900 software companies for $440 billion. Those portfolios are now under pressure. UBS estimates 25-35% of private credit is exposed to AI disruption risk. Arcmont and Hayfin have hired consultants to audit their software portfolios. Apollo quietly cut direct lending software exposure from roughly 20% to 10% in 2025.

And it's showing up in deal activity. Bain's annual M&A report revealed that 1 in 5 dealmakers walked away from a deal in 2025 specifically because of anticipated AI impact.

I've been talking about this for a while. I wrote a blog post back in October arguing that the majority of legacy software we use today will be made obsolete within the next five years. The reasoning is threefold.

First, most software actually kind of sucks. It's rigid. It enforces workflows upon the user rather than adapting to how you actually work. We accepted it because there wasn't an alternative. Now there is.

Second, the wonderful thing about AI is that it allows you to go much deeper on workflows. But the deeper you go, the more specific they become to each firm. So software needs to become more customised and more user-specific with better, increased functionality. That's the opposite of how legacy platforms are built and just compounds problem one.

Third, all of these existing tools are running on codebases that are 25 years old, connected to millions of users who rely on them and for who any disruption could be catestrophic. They're not set up for agents. What do I mean by that? We've spoken about this recently, but software in 2026 needs to be built in a way that makes it easy for agents to access information, understand the context of where it came from, and read and write back easily. The human is obviously still important. But it's a firm belief of ours that over the next few years, we're going to be in a world where multiple agents are pulling together your analyses, and you're there to review, approve, and direct.

That's what we've been building at DealSage. Every functionality is writable as a tool for an agent. That's not a nice-to-have anymore. It's the whole game. I don’t doubt these large players are attempting to refactor their code bases but it will be a long and painful process.

I do want to push back on one narrative that's gaining steam. The idea that every enterprise is going to "vibe code" their own software is nonsense. That's not the risk to these incumbent firms.

Think about it practically. Average IT spend is about 5-8% of revenue depending on the industry. Software is a fraction of that. If you're a PE firm with a billion-dollar fund to deploy, your time should be spent on how to deploy that capital and drive value within your portfolio. Not trying to whittle down 1-2% of your total spending by building your own tools.

And even if you could build something, you'd then have to maintain it. Patch it. Keep it secure. Stay current as the models evolve every few months. That's a full-time job. Most firms don't have the engineering talent for it, and hiring that talent is expensive and competitive. It's just not viable for the vast majority of organisations.

The real risk to incumbents is that small teams with modern architectures can deliver a more personalised experience. That's what's going to replace these monolithic businesses. Not enterprises building their own tools. Startups building better ones.

My strong sense is that outside of the frontier model providers, Anthropic, OpenAI, and Google, which are going to be enormous, the rest of the software market is going to be comprised of much smaller companies serving specific niches. Deeply specialised. Built for agents from the ground up. Serving workflows that the big platforms can't reach with the level of care and personalisation required.

Anyone who's been reading this newsletter from the start will be familiar with my ongoing gripes with how terrible Microsoft’s Copilot is.

Months later, the Wall Street Journal caught up. The headline: "Microsoft's Pivotal AI Product Is Running Into Big Problems."

Now look, Microsoft is probably going to be fine. They have large stakes in all the major frontier models. The world runs on their software. But for anyone who's used Copilot, it's symptomatic of exactly the challenges I've just described. Microsoft has 450 million paid M365 seats. Copilot penetration? About 3%. Roughly 15 million paid users. Recon Analytics found the percentage choosing Copilot as their primary AI tool fell from 18.8% to 11.5% between July 2025 and January 2026. A Citigroup survey found companies using only about 10% of their Copilot quotas.

Nadella himself sent a frustrated email to his EVP after trying to use Copilot and getting nothing useful back. Microsoft spent over $60 million on TV ads for Copilot in 2025. They're now pulling Copilot integrations back from Windows apps after user backlash. It's really hard to retroactively fit this stuff into legacy providers. And according to the WSJ, it seems like everybody else agrees.

The immediate market reaction might have been somewhat overblown. But long-term? The world of software is going to look markedly different in five years. And the majority of legacy providers aren't going to be able to adapt fast enough to protect their positions.

News Digest

Goldman Sachs Partners with Anthropic to Automate Back-Office Operations

Goldman Sachs revealed on February 6 that Anthropic engineers have been embedded inside the bank for the past six months, co-developing autonomous AI agents for trade accounting and client onboarding. CIO Marco Argenti described the agents as "a digital co-worker for many of the professions within the firm that are scaled, are complex and very process intensive."

The discovery process started with coding. Goldman had already deployed Devin, an autonomous coding agent, across its engineers. Then they asked a bigger question: "Is Claude good at coding because coding is special, or because the model can reason through complex problems step by step?" The answer was the latter. And executives were "surprised" at how well it translated to compliance and accounting work.

Why it matters: If Claude can pass Goldman's risk and compliance teams for accounting and KYC, that validates its readiness for any regulated industry. Healthcare, insurance, legal. And the Fieldguide investment the same week tells you Goldman is building a thesis, not running a pilot.

My take: Two things stand out. First, Goldman embedded Anthropic engineers inside the bank for six months. That's not a software deployment. That's implementation. You cannot overstate how important it is to have trusted partners working alongside your people on this stuff. It's not straightforward and it's not something you just roll out. If Goldman needs six months of embedded engineers to get this working, what does that tell you about firms trying to do it with a chatbot and a prayer?

Second, they started with coding. The lowest-risk, most measurable use case. Got the results. Then expanded into accounting and compliance. That's the pattern. Start where you can prove it, then move into harder territory. The firms trying to jump straight to "AI does diligence" without building that foundation are the ones who'll struggle.

Read more

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

Wells Fargo hires AWS exec as Head of AI Products (Jan 26) - Faraz Shafiq joins from Amazon Web Services, reporting to Saul Van Beurden. Another megabank appointing dedicated AI leadership. That's a luxury mid-market firms don't have. They'll need trusted partners instead.

JPMorgan: 65% of family offices want to invest in AI (Feb 2) - Survey of 333 family offices with average $1.6B net worth. But 57% have no exposure to VC or growth equity where the innovation actually lives. 79% have zero infrastructure allocation. Everyone wants the AI theme. Nobody's positioned for it.

Palmer Luckey's Erebor Bank gets national charter (Feb 6) - First new bank charter under Trump 2.0. $635M in capital, $4B valuation, backed by Thiel, Lonsdale, a16z. Plans to serve AI, crypto, and defence companies. Named after a Tolkien fortress. Because of course it is.

OpenAI launches Frontier enterprise platform (Feb 4) - $10K+/month for enterprises wanting custom agent capabilities. Because charging $200/month wasn't enough.

Goldman leads $75M Series C in Fieldguide at $700M valuation (Feb 2) - AI-native audit platform used by half of Top 100 US accounting firms. CPA exam candidates at 17-year low. Same week as the Anthropic partnership. Goldman isn't dabbling. They're building a thesis.

Anthropic launches Claude Opus 4.6 with agent teams (Feb 6) - New model with 1M token context window and multi-agent coordination. This is the model powering the Goldman partnership. When the model that just wiped $300B from software stocks gets an upgrade, pay attention.

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