I talked to the CPO of a 60-person product team that is on the frontier of AI-enabled product work. This is a snapshot of their current workflow, including what's working but also what needs work.
A human decides a business problem is worth solving, then describes and defines it to get the process going.
What used to be the executive summary at the top of a PRD is now a context prompt — the same information, reframed as input for an agent rather than a document for humans.
Humans and agents form hypotheses on the causes of the business problem, narrow to the likeliest candidates, then brainstorm and narrow to potential solutions worth prototyping.
Agents are hooked into the full company context — sales calls, CS calls, prior UXR, customer feedback, tickets, past PRDs. Humans still gut-check with the field.
Replaces the long, manual research-and-synthesis phase. Judgment stays human; the surface area of available information expands by an order of magnitude.
PM, design, and eng form a pod to build working prototypes together — no PRD, no handoffs. All three disciplines work simultaneously around a shared agent.
Bringing all three disciplines together at the prototyping stage eliminates handoffs. Eng joins early to get acquainted with the problem before it's fully defined — and to build prototypes that can move to production faster.
User research against the prototype before productionization begins.
Not a new step — but it happens faster and earlier because of rapid prototyping.
For experiences that pass validation, ~10% of design work remains — a combination of manual fixes in Figma and wrestling with the AI.
The net effect on designer job satisfaction is an open question. More prototypes, more variety — but more slop cleanup too.
For this team, the downstream QA step was a mess for a while. The reason is that the team stopped creating proper test cases at the same time they stopped creating user stories and acceptance criteria (thanks to prototyping). This company has temporarily mended their QA situation by assigning an agent to generate test cases for traditional manual QA.
The agent generates test cases from a running list of past decisions that were made during the planning and prototyping phases.
Engineering team takes prototypes (which are ideally closer to production-ready than not) and does last mile work to hook it into the larger code base and get it ready for deployment.
There is meaningful upfront work for eng and product teams to reliably create prototypes that can go into production quickly. But when it starts to work, it does feel magical.
Standard QA.
While QA is a process that seems ripe for complete AI disruption, at this stage, this company has mostly kept the QA process intact.
Team moving toward automating test setup and experience design, but humans still in the mix.
One promise of agentic development is infinite A/B testing. One real constraint: not enough site traffic to run all desired tests simultaneously.
Full rollout to all users.
The CPO noted a backlog of production-ready code that hasn't shipped — not a technical bottleneck, but a GTM one. The sales and CS teams can't be trained on new features fast enough to keep pace with how fast the product team can now build.