PM writing and AI workflows

Product managers are rewriting the way they write as AI products move from single prompts to systems that route work across multiple agents. (openai.com) (anthropic.com) In these systems, one model often plans, another searches, and another reviews, so the product spec has to spell out who does what and when control changes hands. OpenAI’s Agents software development kit documents “handoffs” as a built-in way for one agent to delegate to another. (openai.com) (openai.github.io) That is showing up in public posts from product managers in April 2026, where Adaline, Jacelyn Smitth, and Khilan Haria described newer spec questions around agent roles, handoffs, and what the user should see versus what stays in the background. Their posts also described lighter documentation and more day-to-day operating work around AI features. (x.com 1) (x.com 2) (x.com 3) The shift tracks the way AI companies now describe production systems. Anthropic said its Research feature uses a lead agent that plans the research process and launches parallel agents to search simultaneously, turning what looks like one assistant into a coordinated workflow. (anthropic.com) Once a product behaves like a workflow, the old product requirement document question of “what should the feature do?” expands into “which actor does each step, what can fail, and when does a human step in.” OpenAI’s documentation says agentic applications use tools, specialized agents, partial results, and traces of what happened. (openai.com) That changes what product managers have to define for users. If an agent is searching, calling tools, or transferring work to another agent, teams have to decide which steps are visible in the interface, which steps need approval, and which steps should stay hidden to avoid clutter or confusion. (openai.com) (openai.github.io) It also blurs older boundaries between product, design, and engineering. Productboard said in January 2026 that it was launching Spark, an “agentic” product management tool that creates briefs, synthesizes feedback, and delivers competitive analysis, reflecting how product work is being pulled closer to live systems and operational loops. (productboard.com) The lighter-docs trend fits that pace. When teams are shipping AI features that change weekly, long static documents age fast, and shorter artifacts can be updated as prompts, tools, and routing logic change. (anthropic.com) (openai.com) The result is not that writing disappears. It is that product writing moves closer to service design: define the roles, define the handoff, define the fallback, and make clear what the user can trust the system to do on its own. (openai.github.io) (anthropic.com)