How Candidly uses Notion to make Cursor agents multiplayer and ship 73% more value

One of Candidly’s upcoming projects is a public benchmark to measure different agents’ ability to take financial actions—like canceling a subscription, finding a better loan offer, or transferring funds safely.

They’re using a Notion task board to manage Cursor agents in a multiplayer environment: product, engineering, and coding agents can see the same tasks, context, and decisions as the work moves forward. Because this particular project is complex and high-stakes, every task needs clear criteria, careful sequencing, and human visibility.

Each ticket links the objective and acceptance criteria to the PRD, decisions, meeting notes, and product documentation behind it. Ben can then launch a Cursor coding agent with the same context the team uses, so it understands both what to build and why.

From one board, Ben can:

• See which tasks are ready, active, blocked, under review, or complete

• Inspect an agent’s progress and respond inside the ticket

• Pull in a colleague without losing the project history

• Step in when the work needs a product decision

Building this way has reduced the time it takes to deliver a full user story by about 10 days, making development 41% more efficient. Now they’re spending more time deciding what to build and QAing code—in other words, making sure what’s delivered is valuable.

Candidly is taking the idea further by testing agents that manage parts of the board itself. They can evaluate finished work, advance a dependent ticket, and involve Ben only when judgment is required, so work continues overnight without becoming invisible.

Previously, PMs had to translate code changes into product explanations and remember every internal guide that might need an update. Candidly wanted documentation to begin with what the product actually does, not relying on a busy person’s memory.

During a three-day sprint, Ben created “code-based treasure maps”: markdown guides that point agents to the files, services, and concepts behind each product area. This sets up agents to navigate the codebase efficiently.

Then, PMs used Cursor, Claude Code, and Codex to draft a consistent product library directly from the code, covering how each capability works, what different teams need to know, and what changed recently.

Two people independently reviewed every initial page before it became canonical. Now, agents built with LangSmith Fleet inspect every merge to develop, identify affected pages, and propose section-by-section edits. A document owner approves, rejects, or changes each proposal before it becomes the new source of truth.

PMs spend roughly 90 minutes per week reviewing documentation work, down from about 10 hours spent repeatedly reconstructing and rewriting it.

The validated library gives every team a faster path to the product truth. Sales can prepare for a customer conversation, Customer Success can check program behavior, and Support can trace an answer back to its source—without waiting for a PM. The same answer can become customer-facing material, like a Candidly-branded slide, without another multi-team handoff.

The speed comes with accountability: every answer points back to maintained, reviewable documentation with a clear owner. PMs can focus on the roadmap and material changes instead of routing every product question.

Ben combined planning data, Jira tickets, and GitHub commits to measure customer value delivered over time. His analysis found that Candidly’s EPD organization shipped 73% more value per employee after adopting agents—even as the team became more than 25% smaller. This was not more code for code’s sake; it was more meaningful work reaching customers.

The model also makes previously impractical ideas possible. One effort to make Cait’s calculations more verifiable became a published research paper and influenced Candidly’s approach to user guidance. The public benchmark for financial-action agents is another multi-month idea Ben can now advance asynchronously instead of leaving it on a “someday” list.

Candidly has generated more than $2.5 billion in impact on users’ finances. For Ben, that mission is the point: Less time maintaining documentation and coordinating routine work means more capacity to improve Cait and help people move from debt toward wealth.

Next, Candidly wants to make more projects multiplayer—bringing Product, Design, Engineering, and agents into the same visible project state. People set direction and make judgment calls; agents handle bounded implementation and maintenance work. Decisions, context, and results stay connected as the work moves forward.