Market Pulse
Every major platform declared itself the agent control plane this week. The competition that matters isn't technical. It's about where governance decisions crystallize.
Market Pulse
Every major platform declared itself the agent control plane this week. The competition that matters isn't technical. It's about where governance decisions crystallize.
The Governance Layer Land Grab
In a three-week stretch, Google, Cloudflare, Salesforce, Databricks, Snowflake, and Microsoft each announced an agent governance platform. Every one of them called it a "centralized control plane." Six centralized control planes is, of course, zero centralized control planes. But the timing is the tell.
These companies don't coordinate product launches. They do read the same enterprise buying signals. When six vendors converge independently on the same claim from six different layers of the stack, the teams caught between six competing definitions of what "governed" means have the harder problem.
The Governance Layer Land Grab
In a three-week stretch, Google, Cloudflare, Salesforce, Databricks, Snowflake, and Microsoft each announced an agent governance platform. Every one of them called it a "centralized control plane." Six centralized control planes is, of course, zero centralized control planes. But the timing is the tell.
These companies don't coordinate product launches. They do read the same enterprise buying signals. When six vendors converge independently on the same claim from six different layers of the stack, the teams caught between six competing definitions of what "governed" means have the harder problem.
Research Signals
TrajAD: Trajectory Anomaly Detection for Trustworthy LLM Agents
Frontier LLMs can't reliably locate errors inside agent trajectories. A purpose-built 4B model beats them handily, suggesting auditing is its own discipline.
Trajectory-level auditing lives naturally in the observability layer. That makes it a control plane function, not a debugging convenience.
Research Signals
Ontology-Constrained Neural Reasoning in Enterprise Agentic Systems
Exactly where models know the least. Compliance and role consistency gains were largest in data-sparse regulated domains.
If ontological constraints shape agent decisions before they execute, external enforcement becomes a backstop rather than the whole story.
Research Signals
Stanford AI Index 2026: The Jagged Frontier
Models that dominate competition-level math perform at coin-flip accuracy on analog clocks. Competence boundaries are invisible from inside the system.
That we can characterize where systems fail. The jagged frontier suggests those failure boundaries shift faster than any framework can track.
Research Signals
Don't Retrieve, Navigate: Distilling Enterprise Knowledge into Navigable Agent Skills
They get a bird's-eye view of the knowledge corpus and choose where to look, rather than consuming search results passively.
The knowledge hierarchy is a designed artifact with editorial authority. Ownership of it is a governance decision hiding in plain sight.
The Headless Turn
Parker Harris posed the question before TDX: "Why should you ever log into Salesforce again?" He wasn't being provocative. He was describing the roadmap.
Headless 360 exposes every capability as an API, MCP tool, or CLI command. Sixty-plus MCP tools shipped day one. A platform that spent twenty-seven years pulling users into browser UI just re-optimized itself for agents that will never open one.
Salesforce is wagering that its accumulated context, the escalations, renewal timelines, relationship graphs, gains value once agents can reach it without navigating screens. Strip the interface and what remains is data gravity. Every incumbent platform should be watching closely.
Worth Reading
Past Articles
A purchasing card knows what you can't buy. Wrong merchant category, declined at the register. Over the spending limit, ...
A system wins gold at the International Mathematical Olympiad, working through proofs that stump most graduate students....
Over 30,000 exposed instances cataloged. Hundreds of malicious skills traced to a single threat actor. CVEs scored and p...
Judge Rita Lin's preliminary injunction in the Anthropic-Pentagon case runs forty-three pages. It is careful, narrow, an...