From document to prototype, without switching apps
On April 13, Atlassian launched three third-party AI agent integrations in Confluence: Lovable, Replit, and Gamma.
The logic is straightforward: you write a product requirements document in Confluence, open Rovo Chat, and ask Lovable to turn it into a runnable UI prototype. You don't need to manually copy and paste content. The agent reads the Confluence page directly, including metadata such as the document's author, associated project, and historical decisions, and brings it all into the target tool.
Replit does the same thing, but for code: it turns a technical document into a starter project that engineers can fork. Gamma turns documents into presentation slides.
All three agents are built on the MCP protocol.
What each agent does
| Agent | Input | Output |
|---|---|---|
| Lovable | Product requirements document | Runnable UI prototype |
| Replit | Technical document | Forkable starter code |
| Gamma | Any document/report | Presentation slides |
The trigger is the same for all: open Rovo Chat within a Confluence page, enter a command, and the agent reads the current page and its context (author, project affiliation, related decisions), then generates the corresponding content in the target platform. No need to re-explain the background or paste documents.
Remix: Let AI handle visualization
Atlassian also launched its own Remix tool, now in open beta. Remix converts data from Confluence pages into visual content — charts, infographics, scorecards.
Select some content, tell it what format you want, and it generates it for you without leaving Confluence to open another application.
Tools like this aren't new, but embedding them into an existing workflow rather than creating another interface is a design direction worth noting.
Context: 1,600 layoffs a month ago
These two events are worth looking at together.
In March, Atlassian announced it was laying off 1,600 people, about 5% of its workforce. At the same time, the company said it would invest more resources into AI feature development.
This isn't a script unique to Atlassian: layoffs → AI feature launches — this sequence has become almost standard in the software industry. Microsoft, Salesforce, and Google have all followed similar timelines.
The cost of paying employees is being converted into the cost of purchasing API calls. The logic is clear for the company; it's a different story for those laid off.
MCP in the enterprise: A different direction
Recently, discussions about the MCP protocol have mostly revolved around "letting AI connect to external data" — Claude connecting to databases, connecting to search tools.
Atlassian's case flips that direction.
MCP isn't about letting AI read external data; it's about letting external AI agents read internal Confluence documents and then go work in other tools. Confluence has shifted from a document repository to a context source for AI agents.
This means years of accumulated internal documents — product specs, technical designs, decision records — can, for the first time, be directly used by automated workflows, not just manually searched by people.
If this model extends across the entire enterprise software ecosystem, with data in Jira, Notion, and Salesforce all accessible to agents, it changes how deeply AI can operate within an organization.
Pinterest: 66,000 MCP calls per month; Atlassian: 1,200 at the summit
The MCP protocol has spread rapidly since late last year. Pinterest runs MCP in production, with over 66,000 calls per month, saving 7,000 hours of manual processing time. Last month's MCP summit drew 1,200 attendees, and Uber reported running over 10,000 agent tasks per week.
Atlassian's integration is another data point: this is no longer startups testing new technology; it's mainstream enterprise software companies embedding it into their products.
Still early
The three agents launched this time cover UI design, code, and presentations — clearly targeting developers and product teams.
For Confluence's main user base — regular enterprise employees who write weekly reports and store meeting notes — this is just a small opening.
Atlassian says MCP is a standard protocol, and theoretically any tool that supports MCP can be integrated. Integration capability and actual coverage are two different things. How many agents will come in and how broad the coverage will be are the real questions.
Three agents, today. Where it goes from here will be determined by the data.
Sources: Atlassian launches visual AI tools and third-party agents in Confluence (TechCrunch); Atlassian brings AI visual tools and partner agents to Confluence, 1 month after cutting 1,600 jobs (The Next Web); CocoLoop, Atlassian Introduces Visual AI Tools and Agents in Confluence to Transform Team Collaboration (CXO Digitalpulse)