← All work signals
Trust signal

The Agent Divide

McKinsey’s survey has 40% of large companies scaling AI agents in a function, against 22% of smaller firms. Scaling an agent is not the same as letting it run alone.

What is it?

A near nine-billion-dollar category has formed around autonomous agents in finance, IT and procurement. In McKinsey’s survey, 40% of large companies scale agents in at least one function, against 22% of smaller ones, but scaling is not the same as letting an agent run unsupervised. Whether that gap closes, or is the ceiling, is open.

Why now?

Capability arrived faster than the governance needed to trust it with a live process. Interoperability standards now let agents plug into shared tools, raising the stakes of who answers for an action nobody watched happen. The bottleneck moved from capability to accountability.

Who’s building it?

Enterprise software vendors shipping agent tooling — accountability for what it does still sits above them.

Agent platform vendorsEnterprise IT teamsModel providersInterop standards bodiesCompliance officers
The full read
Why now, who is building it, where the market goes, and what stays unsettled.
Get weekly →
01 · The market
Who’s in it
Builders — vendors wiring agents into finance, IT and procurement workflows.
Buyers — enterprise IT and operations teams piloting agents inside existing systems.
Gatekeepers — compliance and governance teams deciding what earns production trust.
What’s at stake
The value of a nine-billion-dollar category concentrates in the companies large enough to scale agents, and in the share of them trusted to run alone.
Accountability blurs once an agent acts without a person watching.
A pilot that never ships wastes the whole build.
What’s unsettled
Whether governance catches up before the pilot backlog stalls out entirely.
Who answers when an interoperable agent acts across systems.
Whether the 22% at smaller firms closes on the 40% at large ones, or stays a ceiling.
02 · How the demand matureda progression, not a timeline
Experimentation
Teams wire agents into low-stakes tasks just to see what they can do.
Pilot
A scoped process runs supervised, with a person still checking every output.
Production trust
The agent runs unsupervised on a real process, and nobody has to double-check it daily.
Standard practice
Interoperable agents become as unremarkable inside operations as the software they replaced.
03 · Where the market goesthe forks, not a prediction
→Trusted or stalled — decided by whether governance tooling matures fast enough to certify an agent for a live process.
→Interoperability standards could let agents plug into each other faster than any single vendor governs.
→Either scaling spreads past large companies, or smaller firms’ pilots quietly stall at demo stage.
→Accountability could land with the vendor, the deploying team, or nobody clearly at all.
Everything here turns on one premise: that an unsupervised agent can be trusted with a mistake nobody sees happen in real time.
04 · The Agentic Delegation deskadjacent markets, moving now
Who Is This Agent
The identity problem sits under this gap.
Shared API keys hide which agent acted.
GitGuardian
Paying Without a Person
Autonomy extends from decisions into real spend.
Stripe, Visa, Mastercard race to build agent rails.
Forbes
Agentic Commerce
The same delegation logic, applied to checkout.
OpenAI and Stripe launched an open commerce protocol.
CNBC
05 · The Wire
01
Market nears nine billion dollars.
Tech Insider ↗
02
Agents reshape operations, Gartner says.
Itential ↗
03
40% of large firms say they scale AI agents in a function
FM Magazine ↗
The panel on this signal
They agreed on
–The gap between piloting and trusting an agent in production is the real competitive line, not raw adoption numbers.
–Interoperability raises accountability stakes faster than it resolves them.
–Governance tooling, not model capability, is what is actually holding deployment back.
The ML systems engineer disagreed on
–The gap between agent installation and unsupervised production use may reflect a durable governance limit around auditability and error accountability, rather than a temporary tooling shortfall that better monitoring dashboards would resolve.
Know more about the method →
✦How this signal was built
01 · Panel
A Delphi panel of 4 AI analysts, each reasoning independently.
→
02 · Debate
They challenge and revise each other’s reads over two rounds.
→
03 · Evidence
Every claim is tied to a cited, linked source on The Wire.
→
04 · Synthesis
A moderator writes it into one read, disagreement and all.