How to manage productsin the era of AI.
Spoiler: we still do not have all the answers —
but some things have definitely changed.
Roma Luev · Notiky · Cursor Ambassador
Raise your handif —
- 1
you use AI every day at work
- 2
your team already has AI agents working
- 3
some part of you is afraid you’ll be replaced
Okay — now I know who I’m talking to.
For twenty yearsPMs had the same 4-step job.
- 01
Write a 40-page spec
- 02
Throw it over the wall to devs
- 03
Wait six weeks
- 04
Ship the wrong thing
We called it product management. Mostly it was translation loss.
AI learned to code.Anyone ships in a weekend.
Cursor
Claude Code
Windsurf
Lovable84%
of developers use AI tools
Stack Overflow, 2025
25%
of Linear issues are authored by agents
Linear, 2026
5×
growth in agent-authored work over 3 months
Linear, March 2026
Most of thisis slop.
Agents type faster than any of us. But they don’t decide what to build, they don’t remember why, and they don’t argue with you.
- #1208feat: add dark mode v47agent
- #1209fix: undo changes from PR #1204agent
- #1210feat: rewrite auth (again)agent
- #1211chore: bump depsagent
- #1212feat: add settings panelagent
- #1213refactor: split the thing we just mergedagent
- #1214feat: add admin dashboard no one asked foragent
The bottleneck moved.
yesterday
Typing fast enough.
Turning a decision into code, one keystroke at a time. The PM translated thinking into Jira tickets, devs translated tickets into lines of code.
today
Thinking clearly.
Deciding. Capturing why. Holding context across every agent that shows up to work on your product.
The PM era ended quietly. The product-thinking era is starting right now.
You no longer manage features.You manage a fleet.
The PM used to be a translator between business and code. Now the PM is a director. Agents work in parallel, on different streams and repos. Your job is to hold the vision, context, and priorities.
Assign like a colleague
Tasks go to the agent with the right skill.
The agent executes alone
Enqueue → claim → execute → report. No babysitting.
Skills accumulate
Every solved problem becomes a reusable primitive for the team.
Твои следующие 10 коллег — не люди. И это ок. Они не просят зарплату и не уходят в отпуск.
10–20 потоков в параллель.
Not a single roadmap. Not a single team. Dozens of playbooks pulled by people + agencies + agents — all at once. The PM sees the board, priorities, and blockers.
landing-v4
runningblog-seo-push
runningpricing-test
claimedonboarding-v2
runningads-q2-creative
shippedemail-drip
runningdocs-rewrite
runningchangelog-v3
claimedcommunity-ops
runningvideo-shorts
runningcompetitor-intel
shippedlifecycle-emails
runningreality check
Base44 — $1M ARR in 3 weeks. Solo. With agents.
what is changing
Compan is not 50 people. It is 5 people + 10 agents + 3 agencies.
What a PM does every week — now.
Think deeply.
Minimum 5 rounds. AI that argues with you. Not the first idea, the fifth.
Decisions with memory.
Every decision captured with reason, alternatives, and evidence. The 50th session is 10× sharper than the first.
Context to any agent.
Vision, architecture, PRD, rules. Auto-delivered to Cursor / Claude / Codex via MCP.
This is the new product management. Three moves. Every week.
What it looks like in practice.
- 01 · input
Brain-dump
Voice, text, or links. Unstructured. Exactly as it is.
- 02 · think
Deep session
5–8 rounds. Assumptions, alternatives, and evidence.
- 03 · memory
Decision
Decision + why + what was rejected.
- 04 · output
Context pack
Vision · PRD · Cursor rules · Claude rules — auto-composed.
- 05 · delivery
MCP → agent
Cursor · Claude Code · Codex · the agent that ships next month.
What to unlearn.What to relearn.
unlearn
- ✕40-page specs
- ✕velocity as the main metric
- ✕Jira ticket factory
- ✕“PM = writing tickets”
- ✕feature parity with every competitor
relearn
- →deep thinking
- →decision memory
- →context engineering
- →agent orchestration
- →taste and product sense
Not all at once. One ritual this week. One next week.
PM-translatorилиPM-orchestrator?
You answer that for yourself. By 2027 we’ll see who was right.
Try Notiky — scan.
спасибо. questions — now.