Over the past few months, I’ve found myself repeatedly solving the same problem:
We have great conversations with clients — rich, nuanced, full of intent — and then we lose signal as we try to turn that intent into an executable product plan.
This post outlines a workflow I’m developing to close that gap. It’s not about replacing humans with AI. It’s about building a system that preserves intent, creates alignment, and makes execution safer and faster.
The Core Problem
Most product workflows break in one of three places:
-
Conversation → Documentation
Nuance is lost. Decisions blur with assumptions. -
Documentation → Roadmap
Roadmaps are either too vague to execute or too rigid to survive reality. -
Roadmap → Execution
Tickets drift from original intent, and AI (when used) amplifies the drift.
The goal of this workflow is to reduce entropy at every step.
The High-Level Flow
Here’s the shape of the system:
Client Conversation
↓
Transcript + Notes
↓
AI-assisted Structuring
↓
Google Sheet (Client Review)
↓
Linear (Execution System)
↓
AI-assisted Ticket Execution (Human-in-the-loop)
Each step has a clear purpose, a clear artifact, and a clear owner.
Step 1: Capture the Conversation
We start with tools clients already use:
- Google Meet for calls
- Automatic transcripts
- Optional AI-generated notes (Gemini, etc.)
These raw artifacts are immutable. They are never edited. They serve as the ground truth if questions arise later.
Step 2: Structure the Chaos (with AI)
This is where AI shines — not as a decision-maker, but as a structuring assistant.
Given transcripts and notes, AI helps:
- Identify candidate epics
- Extract user stories
- Propose acceptance criteria
- Flag assumptions vs. confirmed decisions
The output at this stage is explicitly a draft.
AI is allowed to suggest. Humans are required to decide.
Step 3: The Google Sheet as the Trust Layer
Before anything enters an execution system, it goes through a shared Google Sheet.
Why Sheets?
- Clients understand it instantly
- It’s easy to review live or async
- Comments are first-class
- It exports cleanly to CSV
This sheet becomes the alignment artifact.
When a client says, “Yes, this reflects what we agreed to,” we lock it.
That sentence matters.
Step 4: Linear as the Execution Boundary
Once approved, the sheet is exported and imported into Linear via script.
Important rules:
- The flow is one-way (Sheets → Linear)
- Linear is not used for discovery or debate
- Each ticket traces back to a reviewed row in the sheet
Linear becomes a place for doing, not negotiating.
Step 5: AI-Assisted Execution (Safely)
This is where tools like Claude come in — but with constraints.
The model:
- Sees one ticket at a time
- Receives the full context of that ticket
- Produces plans, drafts, or code
- Never moves tickets or closes work on its own
Think of AI here as a junior teammate: helpful, fast, and always reviewed.
Why This Matters
This workflow optimizes for a few non-obvious but critical things:
- Intent preservation
- Client trust
- Auditability
- Safe AI leverage
It’s not the fastest possible system.
It is a system that scales without collapsing under ambiguity.
What This Is (and Isn’t)
This is:
- A repeatable consulting workflow
- A productized way to turn conversations into execution
- A foundation for responsible AI use
This is not:
- Fully autonomous AI
- A replacement for product judgment
- A silver bullet
Closing Thought
AI doesn’t remove the need for rigor — it demands more of it.
The teams that win won’t be the ones who automate the most.
They’ll be the ones who design the best boundaries.
If you’re thinking about similar problems — or building something like this yourself — I’d love to compare notes.