Everyone’s Deploying AI Agents (and No One Can See What They’re Doing)

AI Agents

The Real AI Agent Problem Isn’t Building Them — It’s Seeing What They’re Doing

What happened: A wave of new research landed this week, and it all points the same direction. Dynatrace’s State of Observability 2026 report — a survey of 1,202 technology leaders across 12 countries — found enterprises are deploying AI agents faster than they can monitor, govern, or explain them. Genpact’s poll of 2,000+ executives reached a similar verdict: 79% say agents are already in production, but only 20% apply them broadly and consistently. Meanwhile, Pathlock’s AI Governance Gap Report found 28% of organizations already let agents approve business transactions — and one in four of those has already logged a related incident.

Why it matters: This is the “so what” hiding under a pile of press releases. The hard part of the AI Employee era was never going to be writing the agent — it’s trusting one with a permission that can update a record, approve an invoice, or modify infrastructure. The market is now pricing that exact gap. Kindo raised $30 million (800% revenue growth) for an “AI Operations Center” that maps sanctioned and shadow agents, attributes spend, and lets operators take over a runaway agent. Tel Aviv’s Bay emerged from stealth with $50 million for a runtime enforcer that reviews every agent action in milliseconds and blocks prompt-injection or “purpose drift.” Two separate $50M-class bets on the same thesis, in the same week.

What’s next: Watch for “agent observability” to become its own category the way application performance monitoring did a decade ago. Deloitte’s Alvaro Martin framed it cleanly: the next phase isn’t deploying more agents, it’s “scaling autonomy where it creates value” — and that requires evidence of how agents actually behave. If you’re building agents, the takeaway is practical: plan for the audit trail and the kill switch before you wire in the permissions, not after.

Sources: Dynatrace report via SMBtech · Kindo (Unite.AI) · Bay (TechGig)

LangChain Doubles Its Valuation to $3 Billion

LangChain raised $300 million at a $3 billion valuation, more than double its figure from last October, in a round led by Alphabet’s CapitalG. It’s a sign that the tooling layer for building agents — not just the models — is where serious capital is concentrating. Source (WSJ)

Amplitude’s Wave Agent Compresses Product Cycles to 2 Days

Amplitude says its Wave agent now runs the full loop — insight to deployment to measurement — compressing multi-week sprints into 48-hour cycles, with OpenAI models handling the reasoning layer. It’s a concrete case study of semi-autonomous agents actually shipping product, not just drafting code. Source (Tech Buzz)

AI News

Jensen Huang Resigns as Nvidia CEO Amid Spy Scandal

Nvidia’s board announced Huang’s resignation effective immediately, amid an FBI investigation into his relationship with a woman suspected of spying for China. The stock dropped 4.3% — roughly $500 billion in value — and Trump floated former Apple CEO Tim Cook as a successor; CFO Colette Kress steps in as interim CEO. This is the single most consequential leadership shake-up in the AI hardware world in years, with real implications for the chip supply chain. Source (CBS News)

Norwest Raises a $4 Billion Fund, Its Largest Ever, to Go Deeper on AI

The 65-year-old firm’s 18th fund will target physical AI, voice AI, agentic cybersecurity, and aerospace, alongside consumer and healthcare. Its recent bets on Coval (agent evaluation) and Generalist AI (robot foundation models) show where the checkbook is pointed. Source (TechCrunch)

UK AI Chip Startup Fractile Hits $7 Billion Valuation

Fractile raised $720 million in a Series C backed by the British government, Thrive Capital, and CapitalG — positioning it as one of the UK’s most valuable tech startups. It’s a clear signal that governments are now directly funding domestic alternatives to Nvidia in the AI hardware race. Source (The AI Insider)

Quick plug: Want to build your first AI employee? Start with the briefing: the free 90-minute guide to briefing your first AI employee — the part everyone skips.

https://go.aitokenlabs.com/digest-build

This newsletter? Written by an AI Employee, approved by a human — so our team stays focused on what only humans can do.