Outcome Driven
What belongs on the roadmap after AI finds 493 security issues?
Anthropic found unnecessary exposure, Figma gave admins a new way to manage app access, and Airtable reported a jump in agent use. Each finding calls for a different product decision.
Imagine taking a list of 493 findings into a roadmap meeting. The engineering lead wants to know what needs fixing. The finance lead wants to know what it will cost. A customer wants to know whether their service still works. The number alone answers none of them.
Three recent announcements give product leaders a better agenda for that meeting. One is about what software exposes by default. One is about who gets to connect an app. The third is about whether people come back to an agent after they try it.
What did Anthropic actually count?
Anthropic used Claude to examine 25 open-source projects and reported 493 issues involving the ways software features become reachable. In the group of eleven projects intended for authenticated users, every project had some code reachable earlier than its authors intended. For that group, the researchers said the issues they described would not have been immediately exploitable on their own. Calling the 493 findings “493 breaches” would give a board an entirely different story. Anthropic’s attack-surface research.
Consider an old import option that still ships enabled. The product question is who depends on it today. Ask support for recent cases, the product owner for the promised behavior, and engineering for the supported configuration. If an existing customer relies on the old path, the change may need notice and a migration. If the team can retire it, assign someone to verify both the new default and the customer’s normal import journey.
Who decides which apps can connect?
On November 17, Figma announced that Organization and Enterprise admins would begin getting an allowlist for OAuth apps, including MCP clients, with a place to review requests. That puts a familiar product question in front of design operations. When someone needs a connector for Thursday’s review, who can make a timely decision? Figma’s November 17 release notes.
Give the requester an example of a good business reason and tell them when to expect an answer. During rollout, look at the oldest unanswered request. A five-day wait for a ten-minute design task tells you something a count of approved apps never will.
Did the agent earn a second visit?
Airtable’s engineering team reported a 60% increase in weekly active users after rebuilding Omni. The share of positive ratings also rose from 75% to over 90%. The release combined several changes, including skills for tasks such as large imports. Those figures describe the release as a whole. They leave open which change mattered to which user. Airtable Engineering’s Omni rebuild.
If you run a similar pilot, follow people who used the agent for one named job. Did they use it again the next time that job came up? Ask a few who returned what they trusted it to do, and ask a few who left where they picked the work back up. Both groups can tell you what to build.
Take one decision to the next review
Try this agenda item on a single feature. “We propose changing ___ for ___. The customers who could feel this change are ___. We’ll know it worked when ___. If ___ happens, the owner will revisit the decision on ___.” Bring the current setting, a customer example, and someone who can approve the change. That is enough to turn a headline into a roadmap conversation.