Writing
Things I read, and what I make of them
Two kinds of writing live here: my own reading notes on articles I came across, and an automatically generated daily column, an AI experiment I built myself.
Daily column8 November 2026Treat Deliberate Exclusions as Product Decisions
Product strategy should define which capabilities will remain absent, because deliberate limits can create more value than another round of feature expansion.
Daily column7 November 2026Supplier Momentum Is Not Customer Demand
Product leaders should stop treating funding, competitor launches and lower build costs as evidence that customers will adopt or pay for a product.
Daily column6 November 2026Your Roadmap Needs an Absorption Limit
Roadmaps should limit concurrent change according to the capacity of customers and the wider company to adopt it, not the capacity of engineers to ship it.
Daily column5 November 2026AI Products Need an Authority Ladder, Not an Autonomy Toggle
AI adoption depends on matching an agent’s authority to the consequences of the task, so teams should design gradual delegation rather than aiming directly for full autonomy.
Daily column4 November 2026Every Automated Decision Needs a Receipt
AI products should preserve the provenance of consequential outputs so users and teams can inspect how a decision was produced.
Daily column3 November 2026When Building Gets Cheap, Products Get Harder to Understand
As software production accelerates, product leaders should treat customer comprehension—not engineering capacity—as a primary constraint on the roadmap.
Daily column2 November 2026Customer Access Is Now Production Infrastructure
As building becomes faster and features become easier to copy, direct and repeatable customer access becomes a core delivery capability rather than a separate research activity.
Daily column1 November 2026A Product Expansion Must Earn the Right to Share the Brand
Product leaders should evaluate adjacent features by whether they strengthen the meaning and economics of the core product, not merely whether users will try them.
Daily column30 October 2026Forced Adoption Produces Bad Product Evidence
When users cannot refuse a product or feature without penalty, adoption data measures pressure rather than value and should not guide the roadmap.
Daily column29 October 2026Do Not Automate Away the Work That Builds Judgment
Product leaders should treat junior decision-making opportunities as essential capacity, not inefficient work to be removed by AI.
Daily column6 October 2026Don't Let an AI Assistant Advise and Sell in the Same Voice
Conversational products need an explicit boundary between helping a customer decide and steering that customer toward a commercial outcome.
Daily column5 October 2026AI Output Needs a Named Cost Owner
Companies should assign ownership for the downstream review cost of every AI workflow, not just ownership for producing its output.
Daily column4 October 2026Treat Distribution Access as a Product Dependency
Product leaders should design for the loss of a platform, integration or permission instead of treating continued access as a commercial certainty.
Daily column3 October 2026Measure Whether Customers Recover When Your Product Fails
Growth teams should treat recovery from errors, denied permissions and interrupted tasks as a measurable customer journey rather than an edge case for support.
Daily column2 October 2026AI Makes Product Removal a Core Team Capability
As AI lowers the cost of adding product behavior, teams must become equally capable of consolidating and removing it.
Daily column1 October 2026Build Your Product Operating Model Around Coordination Costs
Product leaders should design roles and rituals around where their own organization loses time and context, rather than importing a fashionable operating model.
Daily column20 September 2026Your Product Backlog Is Hiding Unfinished Decisions
Product leaders should manage unresolved decisions with the same discipline as delivery work, including a deadline, an owner and a default outcome.
Daily column19 September 2026Evidence Quality Belongs in the Product Budget
Product leaders should fund the systems that make results trustworthy, rather than treating measurement as a free by-product of shipping.
Daily column18 September 2026Discovery Is Incomplete Until Operations Can Survive the Product
Product discovery should test whether the company can operate a product at scale, not only whether customers want it and engineers can build it.
Daily column17 September 2026Staff Product Teams Around Failure Modes
Team design should follow the ways a product can fail, rather than mirroring its feature areas or technical architecture.
Daily column16 September 2026Usage Can Validate Demand and Still Damage the Business
Product teams should treat marginal cost and gross margin as part of product validation, rather than discovering after growth that popular usage is economically harmful.
Daily column15 September 2026Your Product Must Earn the Context Switch
As platforms absorb more adjacent tasks, standalone products should compete on whether their added value justifies asking users to leave the environment they are already in.
Daily column14 September 2026AI Agents Need an Authority Map, Not Just Permissions
Treating agent access as a technical permission problem hides the business decision of who may commit money, customer trust and operational capacity.
Daily column13 September 2026Product Teams Need a Formal Right to Dissent
Product autonomy only works when teams have a defined way to record, escalate and revisit disagreement without turning every decision into a consensus exercise.
Daily column12 September 2026Your Roadmap Has an Assumption Limit
A roadmap becomes unreliable when one initiative depends on too many unproven assumptions succeeding at the same time.
Daily column11 September 2026Not All Friction Is a Conversion Problem
Product teams should distinguish accidental friction from deliberate friction that protects capacity, service quality and customer intent.
Daily column10 September 2026Ambient Products Need Boundaries, Not Engagement Goals
As AI products move into the background, product teams must stop treating engagement as the default sign of value and start managing observation, interruption and operating costs.
Daily column9 September 2026When Building Gets Faster, Review Becomes the Bottleneck
AI does not remove delivery constraints; it moves them from producing work to validating, integrating and operating it.
Daily column8 September 2026Internal Vocabulary Is Leaking Into Your Product
Product leaders should treat inconsistent terminology as product debt because it creates friction across acquisition, onboarding, support and measurement.
Daily column7 September 2026Manual Mode Is a Core Feature, Not a Failure State
As automation takes on consequential work, product teams should design user takeover and recovery as part of the primary experience rather than as an emergency fallback.
Daily column6 September 2026Product Leaders Must Own the Claims Their Products Make
Product management should treat every customer-facing promise as part of the product, with explicit evidence, scope and ownership.
Daily column5 September 2026A Pilot Without a Stop Condition Is Already in Production
Product teams should define when and how to halt a pilot before agreeing on the evidence required to expand it.
Daily column4 September 2026The Default Customer Hides Your Best Product Decisions
Product teams should treat differences in language, regulation, workflow and trust as primary discovery material rather than localization work at the end.
Daily column3 September 2026Every Critical Integration Needs a Replacement Path
Product leaders should treat the cost of replacing a critical vendor or platform as a product metric, not an architectural detail.
Daily column2 September 2026A Blank Prompt Is a Product Team Avoiding a Decision
AI products should replace open-ended prompt boxes with opinionated starting points that encode the choices users hired the product to make.
Daily column1 September 2026Measure the Time From Evidence to Decision
Most product organizations diagnose slow delivery when the real delay sits between receiving credible evidence and making an explicit decision.
Daily column31 August 2026Audience Overlap Is Not a Product Expansion Strategy
Companies should expand only when an existing operating capability gives them an unfair advantage in solving the next problem, not simply because current users might want more things.
Daily column30 August 2026Define Product Roles by Decision Rights, Not Task Lists
Product organizations should clarify ownership by naming who makes which decisions, rather than assigning activities to roles.
Daily column29 August 2026A Roadmap Is Also a Capital Allocation Plan
Product leaders should evaluate roadmap choices by the capital they commit and the payback they require, not only by user value and delivery effort.
Daily column28 August 2026Roadmap the Permission Ladder Before the AI Agent
Agentic products should be planned as a sequence of permissions, not a sequence of increasingly impressive capabilities.
Daily column27 August 2026When Output Gets Cheap, Product Teams Must Measure Removal
As AI increases the volume of code, ideas and analysis, product leaders should treat removal as a core operating discipline rather than rewarding teams for producing more.
Daily column26 August 2026AI Memory Needs an Expiration Policy
Persistent AI memory should be treated as a governed product state with explicit ownership, correction and expiry rules, not as a convenience feature.
Daily column25 August 2026AI Business Cases Must Price the Supervision
AI investments should be judged on the full cost of producing an accepted outcome, including human review, exception handling, security and maintenance.
Daily column22 August 2026The Model Demo Is Not the Product Review
As AI performance shifts from the model to the surrounding system, product reviews must evaluate the whole operating loop rather than celebrate isolated output quality.
Daily column21 August 2026Roll Out by Failure Boundary, Not User Count
Product rollouts should expand through complete operating contexts where consequences can be observed, rather than through arbitrary percentages of users.
Daily column20 August 2026Every AI Roadmap Needs a Bundling Test
AI product teams should test whether a larger platform can bundle their planned feature before treating it as a viable product or growth bet.
Daily column18 August 2026A Feed Is an Incentive System, Not Just an Interface
Distribution mechanics should be managed as product strategy because they change user and supplier behaviour long before a roadmap can correct the consequences.
Daily column17 August 2026Usage Is Not Adoption Until It Can Survive Disclosure
AI usage that employees feel compelled to hide is evidence of utility, but not yet evidence of durable product adoption.
Daily column16 August 2026Provenance Belongs in the Product Workflow
AI provenance should be designed as part of the user workflow rather than treated as a technical marker attached to the final output.
Daily column15 August 2026B2B Discovery Must Validate Buyer, User and Operator
B2B discovery must separately validate the needs of the buyer, user and operational owner before a product earns significant roadmap investment.
Daily column12 August 2026Do Not Remove the Constraint That Defines Your Product
When growth slows, removing a product’s signature constraint may improve short-term activity while weakening the product’s reason to exist.
Daily column10 August 2026Product Managers Should Design Decisions, Not Own Them All
The product manager’s real responsibility is to design clear decision rights, not to become the approval point for every product choice.
Daily column9 August 2026Internal AI Spending Needs a Kill Date
The discipline missing from internal AI spending is not better ROI reporting but the willingness to define in advance when an initiative should be stopped.
Daily column8 August 2026AI Makes Delivery Faster. It Also Moves the Bottleneck.
As AI accelerates production, the main constraint shifts from building features to deciding what deserves to be built and what should be stopped.
Daily column7 August 2026A New Interface Needs New Success Metrics
Replacing a familiar interface only creates a genuinely different product when the team also replaces the metrics and assumptions that shaped the old one.
Daily column7 August 2026Good Products Remove Work, Not Judgment
AI makes it easier to eliminate user effort, but product teams must distinguish wasteful friction from valuable acts of judgment.
Daily column7 August 2026AI Agents Need Recoverability Before Autonomy
The most important product requirement for AI agents is not autonomy but recoverability.
- Reading note5 August 2026medium.com
Spec-Driven Development Reframes an Old Requirements Problem
As most AI trend followers know, spec-driven development is the latest trend in the overall AI software-building hype. Personally, I see very similar strengths and weaknesses to the 'old days', when product owners had to be great at writing requirements in Jira issues, most often as user stories, but nowadays teams decide on a common language that works for the team. When reading the article, it strikes me as trying to defend why spec-driven development is an improvement. That in itself already makes me question that notion.