Read Asana’s category announcement quickly and it sounds like marketing. “Agentic work management” is a new label, and new labels from incumbents are usually a way to charge for what they already had. Read it slowly and the pitch is something else. The headline feature is not autonomy. It is a log: every agent action recorded with the access it used, the human who owns it, and what it cost to run. Asana built a category and the load-bearing part is an audit trail.
That choice tells you where the market actually is. We have written before that enterprise agents stall on permissions, not model quality, and that the durable fix is to make the system of record the governance layer. That argument lived in HR and finance, where a wrong action is a paycheck or a restatement. It just walked into work management with a named vendor behind it. The thesis is no longer a prediction. It is a product line.
A New Category Whose Killer Feature Is a Receipt
Asana’s framing rests on a number its own CPO, Arnab Bose, puts at the center: 75% of knowledge workers use AI, but only 5% of organizations report meaningful productivity gains. That spread is the whole problem. Adoption is nearly universal. Value is almost nonexistent. The bottleneck is not whether people will use agents. They already do. It is whether the organization can trust what those agents did enough to let the work count.
So Asana describes four dimensions of human-agent alignment, and it is worth noticing what they are. Work Graph context, so the agent knows what the work means. Multiplayer visibility, so humans see the agent acting in real time. Shared memory, so context does not reset between turns. And an audit trail that logs access, ownership, and execution cost. Three of those are about making the agent useful. The fourth is about making it accountable. The fourth is the one you cannot buy from a wrapper.
This is the same move we flagged when we argued that whoever owns the interface owns control. Asana owns an 18-year Work Graph: the live map of who does what, on which project, under whose ownership. An agent acting through that graph inherits a ready-made record of authority and consequence. A startup bolting an agent onto the same workflow has to reconstruct all of it, badly, from the outside.
Why the Timing Is Forced, Not Chosen
Vendors do not invent governance categories because they feel responsible. They do it when customers stop buying without them. Docker’s first-party data, from its State of Agentic AI work, explains the pressure precisely: 60% of organizations already run agents in production, and 40% name security and compliance as the number one barrier to scaling further.
Sit with the shape of that. The majority are past the experiment. Agents are in production, doing real work, touching real systems. And the single thing stopping them from doing more is not capability. It is the inability to prove the agent stayed inside its authority. A pilot tolerates “we think it behaved.” A production system at scale, audited by people who sign their names to controls, does not. The 40% are not waiting for smarter models. They are waiting for a record they can hand to an auditor.
That is why an incumbent with a system of record can claim a category right now and a model lab cannot. The barrier the market hit is not intelligence. It is evidence. And evidence lives in the system that already knows who was allowed to do what, not in the layer where the conversation happens. Docker’s own definitional work makes the same point from the infrastructure side: governance is not a guardrail you add at the end, it is a property of where authority and logging already sit.
The Audit Trail Is the Product, Not the Feature
Here is the inversion most teams get backwards. They treat the agent’s reasoning as the product and the log as compliance overhead, a tax you pay after the interesting part. Asana’s category bet says the opposite. In an environment where 95% of organizations see no gain from agents they already deployed, the differentiator is not a cleverer agent. It is the ability to answer, cleanly, three questions about any action the agent took: what access did it use, who owns the result, and what did it cost.
Notice that execution cost sits inside the audit trail, not in a separate billing dashboard. That is a deliberate signal. An agent action is not trustworthy just because it was permitted and logged. It also has to be economically legible. A finance leader approving agent work needs the same row to show authority and spend, because an agent that quietly burns budget is its own kind of unguarded action. Putting cost in the audit record treats the agent like an employee whose expenses and access both show up in one place.
Deloitte’s 2026 State of AI work, cited in Docker’s analysis, points at the organizational half of this: companies with strong senior-leadership involvement see significantly greater business value from AI. Read alongside Asana’s 5%, the lesson is blunt. The orgs getting value are not the ones with the best models. They are the ones where leadership owns the operating discipline, and an audit trail is what makes that ownership real instead of rhetorical. You cannot govern what you cannot see, and you cannot lead a system you cannot inspect.
What This Means for How You Evaluate Agent Platforms
If the audit trail is the product, your evaluation criteria change. When a work platform pitches you agents, the demo will be the autonomy: watch it draft the status update, reassign the task, summarize the project. That part is now table stakes, and it is not where the risk lives. The questions that decide whether you can actually deploy are quieter.
Ask where the agent’s permissions come from. If they are inherited live from the platform’s own model of who can do what, you have one source of truth. If they sit in a separate config the vendor maintains, you have a drift problem with a chatbot attached. Ask where the audit trail lives and what it records. A log of prompts and responses is not governance. A log of access, ownership, and cost, anchored to the system that owns the work, is. And ask whether your leadership can read that record without a translator, because the governance gates that block enterprise AI do not open for teams who can only describe what their agents probably did.
This is also the line between running agents and operating them. We have drawn the difference between an AI-only and an AI-first operating loop: the first bolts automation onto a process, the second rebuilds the process so the automation is governed by design. A category built on an audit trail is the AI-first version showing up in a shipping product. The agent does the work. The system of record vouches for it.
Do This Now
Take one workflow where you have agents in production or about to be, the kind that touches something that matters: budgets, customer commitments, headcount, anything a real person owns. Then run a single test against whatever platform hosts it.
Pick one action the agent took last week and try to produce its receipt. What access did it use, on whose authority, and what did it cost to run? If you can pull that from the system that owns the work, in one place, you are operating agents. If you have to reconstruct it from three tools and a guess, you are in the 95% who deployed agents and got nothing they can stand behind. The fix is not a smarter agent. It is choosing the surface where authority, ownership, and cost already live, and refusing to deploy where they do not. Govern at the system of record, or do not call it governed.
This analysis synthesizes Asana’s Agentic Work Management Platform (thelettertwo, June 2026) and What Is AI Governance? (Docker, June 2026).
Victorino Group helps enterprises turn their system of record into a governance layer for agents. Let’s talk.
All articles on The Thinking Wire are written with the assistance of Anthropic's Opus LLM. Each piece goes through multi-agent research to verify facts and surface contradictions, followed by human review and approval before publication. If you find any inaccurate information or wish to contact our editorial team, please reach out at editorial@victorinollc.com . About The Thinking Wire →
If this resonates, let's talk
We help companies implement AI without losing control.
Schedule a Conversation