RSS Amplifier

The AI Agent Economy · Aug 9, 2026

Issue 18 — Trust: the most valuable company in AI won't build a single agent

0
Sign in to vote or save

Atin Agarwal · The AI Agent Economy

Cloudflare is worth more than most of the websites it protects. Verisign runs the invisible registry plumbing behind DNS and is worth tens of billions. CrowdStrike, which secures other companies’ machines rather than selling anything you personally open, is worth well over $100 billion. Palo Alto Networks is worth more than most of the companies whose networks it secures.

Notice the shape: none of these companies builds the thing you actually use, and all of them are worth more than the things that use them. That is not an accident. It is the most reliable pattern in infrastructure — and it is about to repeat, at scale, for agents.

Trust is the third layer of the dependency stack. Once agents act autonomously, the useful question stops being “is this a clever agent?” and becomes “can I rely on what it did — and prove it to someone who wasn’t in the room?” Trust infrastructure is the set of neutral services that answer that: verification, reputation, attestation, audit. It is not part of any single agent. It sits underneath all of them.

Autonomy without trust is just liability with better marketing. The moment an agent takes a real action with financial, legal, or safety consequences — moves money, signs a contract clause, changes a medical record — someone must be able to verify it behaved: before the fact through permissions and limits, after the fact through proof. A human employee comes pre-wrapped in trust infrastructure we forget is there: references, a probation period, a manager, an audit function, an employment contract that assigns liability. Agents arrive with none of it. Every one of those has to be rebuilt as software, and it cannot live inside the agent being judged — a system cannot be its own auditor. It has to sit outside, neutral, watching everything.

Trust infrastructure captures durable value because trust is non-optional at enterprise scale, and the market has proven it over and over. Verisign turned invisible DNS registries into a business worth tens of billions. Cloudflare, which mostly sits between you and sites you never think about, is worth more than most of them — and well above its own valuation a year earlier. CrowdStrike crossed $100 billion selling the assurance that the machines you rely on have not been compromised. Palo Alto Networks is worth more than most of the networks it guards. In every case the pattern is identical: the layer that vouches for the system is worth more than the systems it vouches for, because everyone needs it and no one can safely opt out.

The current trust gap for agents is enormous — there is no standardised verification, attestation, or identity for autonomous actions today. That gap is the opportunity. My dated call, PRED-010: by December 2031, the leading agent trust-infrastructure company will be worth $50 billion or more — more than any single company whose primary product is an agent application. Read that carefully. It says the most valuable company in the agent economy will not have built a single agent. It will have built the thing that lets everyone else’s agents be trusted.

Watch for a reputation or verification service that multiple agent platforms plug into rather than build themselves — the Verisign move. Watch insurers and enterprise buyers starting to ask “who verifies this?” as a procurement gate. Watch funding flow toward attestation and audit startups even while the splashier money chases apps. The trust layer tends to look overlooked right up until it is indispensable.

The consensus bet is on the flashiest agent application — the one with the best demo this quarter. The historical pattern says the boring trust layer underneath outlasts and out-earns every one of them, because applications get replaced and trust compounds. Apps are where the attention is. Trust is where the value settles.

The AI Agent Economy — Why the Next Trillion-Dollar Industry Has No Employees is out now on Amazon and Notion Press. This series expands Chapter 2: the dependency layer.

Related on atin-agarwal.com: all 15 predictions, with their falsification triggers

Read the original on agarwalatin.substack.com

Comments

Nothing yet. Say the first thing.

    Sign in to join the conversation.