Atlassian's design system documentation for Rovo contains an instruction that reads like a koan: clearly identify AI experiences to build trust and transparency. Label it. Make it visible. Show when AI is at work. The product aspiration, meanwhile, is that Rovo becomes so embedded in how teams use Jira and Confluence that nobody thinks about it at all.

The path to invisibility, it turns out, runs through visibility first. And the design docs don't say when to stop labeling.

Rovo crossed 5 million monthly active users by Q2 of fiscal year 2026. Atlassian brands it as "AI teammates," language that saturates product pages, keynotes, and support docs. The teammate framing is deliberate: a colleague who's just there, ambient, part of the furniture. Then in March 2026, Atlassian cut roughly 1,600 jobs, acknowledging with unusual candor that AI "changes the mix of skills we need or the number of roles required in certain areas." Some of the savings fund continued AI investment. The company is reducing its human workforce partly to build something whose highest aspiration is that the remaining humans forget it exists. That tension deserves a moment of silence it will never get.

Rovo is now bundled into all paid cloud plans. Auto-activated. You don't buy it separately, which means you can't cancel it separately, which means you can't easily measure what it's worth on its own. The distribution strategy maximizes adoption at the expense of attribution. If Rovo works perfectly, it becomes part of the texture of using Jira. Nobody writes a renewal justification for texture.

Salesforce is wrestling with the same tension from the pricing side. In roughly eighteen months, Agentforce has been sold three different ways: $2 per conversation, then $0.10 per action, then per-user licenses. The latest move is a flat-fee enterprise agreement for unlimited deployment. Salesforce's president explained the logic:

Price below cost now. Lock in the relationship. Figure out margins later.

This has a precedent. Akamai built brand equity entirely from a product customers never directly experienced, handling roughly two trillion web interactions daily on infrastructure nobody saw. The model worked for years on contractual lock-in and switching costs. It also produced 17 consecutive quarters of delivery revenue decline once the invisible layer commoditized. The counter-strategy, pioneered by competitors, was to build brand loyalty among developers: keep the product invisible where it's consumed, make it visible where the purchasing decision happens.

That bifurcation is probably where agent companies land. Rovo needs to be invisible to the knowledge worker and visible to the Atlassian admin. Agentforce needs to disappear into the CRM workflow while remaining legible to the CIO approving the contract. The buyer and the user are different people, and the product succeeds with one by failing to register with the other.

The infrastructure parallel carries a warning, though. Akamai's margins held as long as the invisible layer remained differentiated. The moment it became commodity plumbing, pricing power evaporated. Agent companies bundling AI into existing platforms are betting that the intelligence layer stays differentiated long enough to justify the investment. The commercial model for software that succeeds by vanishing is, so far, borrowed from infrastructure companies still working it out themselves.