When the Inputs Start to Disappear
2025.13 - 2nd Post - To prepare for a future with fewer TPMs, we need better ways to measure how close that future is so we can start to figure out what is left for us to do, and hone those aspects.
This is the second post in a series where I’m pondering out loud in public about how AI might reshape or perhaps, in my opinion, replace the TPM role.
In the first post, I presented a hypothesis that AI will eventually lead to less TPMs but to measure our trajectory towards that one possible future, we need to change how we frame and measure the work done by a TPM.
I proposed a new rubric which breaks the TPM job down into five core functions: Process, Synthesis, Distribute, Accountability, Troubleshooting.
This post is focused on the first function: Process.
When Inputs Change Because of AI
Process is the part of the TPM job where TPMs take in scattered signals like PRDs, mocks, OKRs, half-written tickets, and hallway chats or in modern times slack/zoom, and many more signals and try to make sense of it all.
But what happens when those inputs stop being messy? Or stop being written by humans at all? Or stop being needed?
We’re already seeing this tectonic shift with tools like:
Replit, Windsurf, Cursor – coding copilots that generate or modify entire features and services
Granola – AI that listens to product convos and syncs decisions to tools like Jira or Linear
Zoom AI, Google Meet AI, Supernormal – auto-summarizing meetings, tracking decisions, and generating follow-ups
These aren’t speculative. They’re already in use. They’re reshaping the work right in front of us. If you are not already leveraging these powerful new tools, you are on a trajectory of obsolescence1 in today’s job market.
So What Happens to the Inputs TPMs Used to Process?
Let’s look at what used to be, what’s changing now with AI, and what that might mean going forward.
By no means is this list conclusive or comprehensive, however it is the most common sources of inputs TPMs deal with in a majority of the organizations out there in the industry.
PRDs & Specs
Before AI:
TPMs review long, inconsistent specs often missing context or clarity or granular details. They chase down PMs and owners for clarity, flagged gaps, and tried to align everyone on what was actually being built. Many times working with engineering teams to make sense of the intent and “what” of the artifact.
This meant long back and forth with product leaders and team over many days, weeks, to create clarity
Now with AI:
Tools like Notion AI, Confluence copilots, and Cursor can generate specs from prompts or chats.
Some teams may even begin skipping PRDs entirely, building from prototypes or AI-generated UIs in Replit or Windsurf and more.
What This Means for TPMs:
Less time ask for edits or formatting specs or making sense of missing pieces chasing down PMs.
Still need to evaluate feasibility, surface misalignments, and ensure shared understanding of what is the objectives.
AI can write the spec or PMs collaborate with AI tools to write consistent, better specs.
🤔 Does this mean TPMs now focus on: “Does this match what we said we’d build?”
Design Specs & Mockups
Before AI:
Design mocks and specs are foundational for customer facing products and services. TPMs worked closely with designers to translate static Figma flows into documentation engineers could build from. Specs filled in the gaps for states, transitions, constraints, open questions.
A lot of time was spent asking:
“Is this screen in scope for V1?”
“What happens on tap?”
“Do we have copy for this error state?”
Now with AI:
We may not need traditional specs as teams and even designers are going straight to very high fidelity prototypes thanks to AI tools:
Cursor and Windsurf allow teams to go from concept to working code directly, skipping traditional handoff artifacts
Figma AI can generate interactive prototypes and annotate flows in real time
AI tools can build interfaces from text prompts or rough sketches, creating production-grade UI that’s editable and testable
The “spec” is embedded in the experience itself
What This Means for TPMs:
Less need to process multi-page design docs.
More time evaluating feasibility, sequencing delivery, and helping teams navigate scope and polish.
The artifact is no longer a PDF or a frame but multiple working prototypes to pick from.
Specs were how we clarified intent and unified teams.
🤔 Now intent is built into the prototype doe this mean our job’s as TPMs is to make sure it can ship and shipping the right prototype?
JIRA / Linear Tickets
Before AI:
Tickets were often vague, incomplete, or redundant. TPMs cleaned them up for planning and cross-team visibility.
Now with AI:
Copilots (like Windsurf or GitHub Copilot) draft tasks and break down work automatically.
Granola listens to meetings and generates context-rich tickets in Jira or Linear.
What This Means for TPMs:
Better-structured tickets, but more volume.
Still need to verify scope, identify gaps, and align execution to intent.
The backlog might manage itself. I dare go further, do we even have a backlog?
🤔 Does this mean TPMs still has to manage the why behind those tickets and make sure it aligns with our overall vision?
OKRs & Strategy Docs
Before AI:
TPMs pieced together priorities from All-Hands, slide decks, and hallway conversations, zoom chats, 1:1s with leaders. Alignment was manual and messy because we need to process and align with wider organizational objectives, team level goals and roadmaps.
Now with AI:
AI can summarize goals, generate alignment docs, and tag workstreams to OKRs.
LLMs translate strategy into execution artifacts.
Tools like Glean can now answer questions based strategy and vision docs + wikis + company artifacts.
What This Means for TPMs:
Faster access to goals, strategy and vision insights.
Still responsible for making sure the strategy turns into coordinated execution across teams
AI shows you the map and can help teams find clarity.
🤔 Does this mean TPMs focus on making sure the path charted at execution level still aligns with strategy?
Meeting Notes & Decision Threads
Before AI:
TPMs juggled note-taking, decision tracking, and post-meeting follow-ups. This meant frantically writing and listening notes often relying on memory or asking for people to repeat what they just said.
Now with AI:
Zoom AI, Supernormal, and Google Meet generate summaries, decisions, and next steps in real time.
Granola can feed summaries directly into tracking tools.
What This Means for TPMs:
Less admin, easier recall.
Still have to catch subtext, hesitation, and what wasn’t said
You know what was said, even next steps and todos are captured.
🤔 Does this mean TPMs are focused on follow-up or momentum?
Slack Threads & Ad Hoc Conversations
Before AI:
TPMs lived in the swirl of Slack, Email, Teams; we are catching decisions, misalignments, and hidden risks in real-time chaos.
Now with AI:
Granola and Slack GPT summarize discussions, extract decisions, and convert chatter into tasks.
AI filters the noise, but also the context.
What This Means for TPMs:
Easier to catch up on things happening in parallel.
Easier to miss the vibe shift when something’s going off-track.
You get the digest and summary. No more spelunking in threads.
🤔 Does this mean TPMs now focus on the intent and in between the lines signals?
So What Are TPMs Still Processing?
Maybe not specs. Maybe not tickets. Maybe not even meetings, in the traditional sense. We now digest beautiful sanitized and normalized, singular crafted summaries that all sound and look the same.
Maybe, what we’re left with is processing alignment, risk, and context, especially the kind AI still can’t grasp, yet.
Emotions. Uncertainty. Competing incentives. The human stuff.
Maybe that is good thing. If AI can clean the artifacts perhaps TPMs still need to sense and focus on the cracks underneath or hidden.
However, there is another side effect of this magical potion called AI that has been gnawing at me…
What Happens to the Next Generation of TPMs? This is the part I don’t have an answer to.
The more structured and self-contained AI-powered workflows become, especially in functional teams with tight scopes and clear ownership, the less need there is for a dedicated TPM at that functional level.
If a single team can move from prompt → prototype → deployment using tools like Cursor, Windsurf, and Slack AI agents, they may not need someone to manage the glue. The glue is becoming more automated. The mess is getting less messy.
If AI replaces the messy, manual inputs we used to learn from, get better at the craft of being TPMs, to build our skills on top off… how do junior TPMs get their street cred, battle scars, that lived experience that is hard to teach?
You don’t get good at this job just by reading summaries. You get good by navigating the ambiguity. By sweating the details in messy projects. By learning how teams really work. You deal with difficult humans to make impossible projects real with scare resources. You soak yourself in the mess because in that Beautiful Messiness of product building is the secret training ground that builds Great TPMs.
If AI removes the mess, how do we develop the sharp instinct? I don’t know. But it worries me.
Next Up: Synthesis
This post focused on how Process is changing what we take in, and how those inputs are becoming cleaner, automated, or in some cases obsolete.
Next up: Synthesis; how we connect inputs, shape narratives, and help teams move forward when things still feel unclear.
Until next time!
-Aadil