Articles
Long-form writing on AI Product Ops, agentic workflows, and building knowledge that both people and agents can reuse. 13 articles published so far.
How Prompt Hotbar added browser-local Like/Love preferences and My Picks without accounts, a backend profile, or public reaction counts.
Finding a useful item once did not make it easier to return to the items that mattered most, but solving that problem with accounts or a backend profile would add complexity before it was needed.
A private local pilot that exposed Prompt Hotbar's reviewed prompt and workflow catalog to Claude Code through a read-only stdio MCP server.
A browser prompt library still requires the operator to leave the active AI client, find an item, copy it, and return. The experiment asked whether the same reviewed catalog could become available inside the AI client without introducing accounts, cloud sync, or write access.
A five-step workflow for inspecting the current state, turning requirements into checks, making the smallest change, verifying with evidence, and stopping when the evidence is unavailable.
AI can move faster than the operator can verify, especially when ‘proceed’ is sent before the current state, checks, and stop conditions are explicit.
A LinkedIn post on why AI adoption does not stay clean after launch, and why teams need source hygiene, skill improvement, output review, team sharing, and lightweight governance.
AI adoption does not stay useful after launch unless sources, skills, outputs, and team-sharing loops have clear ownership.
A practical experiment for the final days of Fable 5 access: give it a project you already consider complete, then ask it to audit both the work and the checklist that approved it.
A stronger model is most useful when it helps expand the verification system, not when it becomes the verification system.
Collections of prompts age badly. Workflows with typed inputs, outputs, and safety notes are what survive contact with real work.
Collections of prompts age badly. Workflows with typed inputs and outputs survive contact with real work.
A portfolio shows what you did. A knowledge product shows what you can be reused for — by people and by agents.
A portfolio shows what you did. A knowledge product shows what you can be reused for.
PiP and background play look like similar convenience features, but they unlock very different user behaviors — and that's why they sit on opposite sides of YouTube's paywall.
PiP and background play look like similar convenience features, but they unlock different user behaviors — which is why YouTube prices them differently.
AI doesn't remove the need for product judgment — it relocates it. The closing entry in a five-part series on how the PM role shifts toward context design, workflow design, and lightweight governance.
AI does not remove the need for product judgment — it changes where that judgment shows up, shifting the PM role toward designing context, standards, and workflow.
Why a repeatable AI workflow isn't automatically a team capability — and what actually makes workflows reusable: clear inputs, stable output shape, shared standards, and components that reinforce each other.
A repeatable AI use case is not automatically a team capability — individual gains stay personal unless the workflow around them becomes reusable.
Four repeatable AI use cases that made a hub/pipeline/governance framework concrete: planning documentation, research support, prototype-oriented workflow, and review checks.
A framework only matters once it shows up in repeatable, connected work — not as impressive one-off AI outputs.
A practical operating model for AI-assisted product planning, built on three layers: Hub (accept the multi-tool reality), Pipeline (standardize deliverables), and Governance (make gains repeatable).
Generating useful AI output isn't the hard part anymore — making those outputs repeatable, connected, and governable across a team is.
AI didn't just make product planning faster — it made misalignment easier to create at scale. Why treating AI adoption as a process redesign, not a tool choice, changes everything.
AI made drafting cheap, which quietly made misalignment cheap to produce at scale — so AI adoption has to be treated as an operating model redesign, not a tool choice.