The AI agent identity land grab hides a deeper problem

I came across Prathima Pinnamaneni's piece at CB Insights on who security companies will buy next in AI agent identity. It's a solid rundown of the recent acquisition spree—nine deals since January, five with disclosed prices, and a handful of startups still in play. But as I read through the valuations and buyer logic, I kept thinking: the article maps the M&A chessboard without showing why the game is being played so frantically.

Here's the part that didn't get enough airtime: those 10,000 undiscovered AI agents inside one Fortune 500 company are not a marketing stat. They're a symptom. A massive, distributed, unquantified liability that existing identity tools were never built to handle.

We track pain signals from practitioners—people actually responsible for keeping these environments sane. Across our database, we see 27,112 distinct problems logged, spanning 92 industries. A huge chunk sits inside identity and access management, with severity scores that hover around 3.6 out of 5. That's not a minor annoyance. That's people losing sleep.

The discovery problem is worse than vendors admit

CB Insights notes that most acquisitions split into two buckets: finding agent identities, and controlling what they can touch. Discovery startups like Astrix, Entro, and Token get bought because enterprises can't even inventory what's running. Access startups like SGNL, Apono, and Keycard get bought because once you know an agent exists, you still can't stop it from doing dumb things.

But our data tells a third story: the problems bleed into each other. When we cluster identity-related pain signals, the most common themes aren't cleanly "discovery" or "access." They're things like:

• "I can't tell which service account created this API key"
• "Our compliance audit found 40% of machine identities are orphaned"
• "We rotated a token and broke three production services"

These aren't separate categories. They're a tangled mess. And that mess is why buyers are paying absurd multiples for startups with tiny revenue. SailPoint paid $200M for Entro, which added less than $3M in ARR. Cyera's deal for Oasis hit a reported $1B against roughly $5M in revenue. Those aren't financial acquisitions. They're panic buys.

What the article misses: the practitioner's burden

Pinnamaneni's analysis stays at the vendor level—who buys whom, what price-to-funding multiple makes sense, who's likely to acquire next. Fair enough for CB Insights' audience. But for seed investors and indie hackers, the real signal is in the unsolved, bottom-up pain. We see 12,221 app ideas generated from problem clusters on our platform. Identity and access management is one of the densest clusters.

When I look at those ideas, a pattern emerges: practitioners don't just want better discovery or better access control. They want a way to see the entire lifecycle of a non-human identity—from creation, through permission changes, to decommissioning. They want to know whose team owns a given agent and what it's actually doing. They want automated evidence for auditors. Nobody's selling that end-to-end today.

That gap is why CrowdStrike is buying multiple adjacent startups in a single week. That's why SailPoint's roadmap explicitly points toward real-time, just-in-time authorization. The big platforms are assembling pieces, but they're years away from an integrated answer.

The market window is still open

CB Insights lists four likely acquisition targets: Token, Defakto, Keycard, and Aembit. They estimate those startups would cost between $135M and $230M each at the current median multiple. That math assumes the multiple holds. It won't.

Oasis is the cautionary tale. It raised $120M at a $700M valuation in March. Six months later, Cyera bought it at the lowest price-to-funding multiple of the five disclosed deals. That suggests private valuations are already out of step with strategic value. Founders who just raised at big numbers might be in for a rude awakening when acquirers start negotiating from a position of "we can build this ourselves."

For investors, that's actually good news. If you're looking at early-stage security startups attacking agent identity, the froth hasn't infected the whole market yet. The massive problem signal—10,000 unknown agents at one company, 27,000+ pain points catalogued—tells me demand will outpace consolidation for years. But the winners won't be feature-level point solutions. They'll be the ones who can stitch together discovery, policy, and enforcement into a workflow that a security team can actually operate.

Where I'd put money

If I were writing a check tomorrow, I'd look for three things:

1. Correlation across identity silos. Startups that can pull data from cloud providers, SaaS apps, CI/CD pipelines, and on-prem directories into one coherent agent map.
2. Policy-as-code with human approval loops. Access control that isn't just automated but auditable by non-engineers. Compliance teams need to understand why an agent had access.
3. Founders who've run security operations, not just built identity tools. The hardest problems are organizational, not technical.

Pinnamaneni's article is a useful snapshot of who's buying what. But the deeper story is that we're still in the first inning of agent identity chaos. The vendor land grab will continue, but the practitioner pain will grow faster. That gap is where the next big company gets built.

Read the original CB Insights piece here.