PlaybookOpen access
A Dashboard Your Monday Actually Needs
One recurring decision, one screen, a freshness stamp taken from the data itself, and a written rule for the first time it disagrees with a spreadsheet.
Resources
Find your next useful thing.
Templates to adapt. Lessons to work through. Ideas to put into practice. Search the whole collection, or start with a topic.
The collection
176 results · Showing 1–18
JournalOpen access
A million tokens of context, and why retrieval still matters
A large context window can help with whole-document work, but input limits, long-context prices, permissions, and retrieval quality still need attention.
JournalOpen access
A model’s cutoff is not a freshness guarantee
A training cutoff is a warning about built-in knowledge, not a guarantee about every answer. Use current sources and check both the facts and the conclusion.
JournalOpen access
A useful voice agent knows how to recover
Natural speech is only one part of a good voice experience. Design interruptions, confirmations, tool failures, and human handoffs before giving an agent a real phone line.
JournalOpen access
A workflow that survives retries: build an intake recovery demo
Run a synthetic intake workflow, interrupt delivery, recover it, and inspect why retries create no duplicate destination records.
JournalOpen access
AI customer service needs a case lifecycle, not just a good answer
Connect AI support to trustworthy customer records, explicit handoffs, and verified resolution—with a case contract you can adapt.
JournalOpen access
AI governance needs a decision record, not another slogan
Build a practical AI use-case inventory, data route, risk register, and review process that stays connected to the system people actually operate.
ChecklistOpen access
AI Pilot Guardrails Checklist
Agree the stopping condition before launch — afterward, the person who would have stopped the pilot is the person who now owns it.
TemplateOpen access
AI Policy on One Page
Organize the policy around data classes rather than tool names, and name the default — silence becomes whatever the person wanted to do anyway.
TemplateOpen access
AI Readiness Interview Script
Ask about the last instance, not the typical one — if nobody can produce the artifact during the call, the process they described does not exist.
FrameworkOpen access
AI use-case scoring sheet
Five scores, multiplied not averaged, with two of them able to veto the result.
ChecklistOpen access
AI Vendor Security Questionnaire
Every vendor answers the encryption questions crisply — watch for the questions they answer in a vaguer register, because that shift is the signal.
JournalOpen access
An AI evaluation should end in a release decision
Connect task-specific tests, calibrated grading, production traces, and incident response into an AI release process that can learn from failure.
JournalOpen access
An AI policy people will actually follow
A practical staff-facing AI policy names approved accounts, data limits, verification rules and escalation. Keep it readable and connect it to the controls behind it.
LessonSeries access
Answer the interview question with a number
Most interview answers are adjectives about teamwork. A four-part structure that makes yours specific, short, and memorable.
FrameworkOpen access
Approval Rules That Actually Approve
If your approver has never rejected one, they are not a control — they are a queue, and everyone downstream now believes a human checked.
LessonSeries access
Ask for data like someone who has to build it
Most data requests get answered twice because the first answer was not what was wanted. Five lines that get you the right thing the first time.
LessonSeries access
Automate the handoff, not the task
Teams automate the step somebody complains about and the total time barely moves. The time was never in the steps — it was in the gaps between them.
TemplateOpen access
Automation ROI Worksheet
Loaded cost, an automation rate you could defend against the last hundred records, and the running cost line everyone leaves blank.
JournalOpen access
Build AI search that can show where its answer came from
Design retrieval around authoritative sources, useful passages, current permissions, and separate tests for search quality and answer quality.
JournalOpen access
Build AI shopping around a trustworthy transaction
A field guide to product data, buyer approval, payment state, and order recovery for conversational shopping systems.
JournalOpen access
Build an AI cost model that survives contact with production
A practical guide to infrastructure choices, workload accounting, and cost per accepted result—with an illustrative budget you can adapt.
JournalOpen access
Build an AI-assisted app you can still own after launch
Go beyond the generated demo with a small build brief, an end-to-end acceptance matrix, and an operating handoff for an application someone can maintain.
BuildOpen brief
Build an automation ROI case
Thirty minutes turning a hunch into a number somebody can argue with.
JournalOpen access
Build automations that can retry without creating a second mess
Design a workflow around completed work, with stable identities, durable state, safe retries, owned exceptions, and a practical failure-test matrix.
LessonSeries access
Build the artifact that beats a resume
A resume claims. An artifact demonstrates. What to build, how small it can be, and why one good one outperforms a portfolio of five.
JournalOpen access
Build the decision, then the dashboard
Connect a metric contract, trustworthy model, accessible report, and decision log so business intelligence changes what happens next.
JournalOpen access
Choose local AI by the job, not the download size
A model that runs on your laptop is only the beginning. Evaluate the whole local workflow: task quality, memory, privacy boundaries, operating effort, and recovery when it fails.
Tool noteOpen access
Claude Project Instructions That Change Behavior
A line in a project instructions file earns its place only if deleting it changes what the model does — here is how to test that.
Tool noteOpen access
Claude: Long Context Still Needs Retrieval
A million-token window is the default now, not a feature flag, which moves the retrieval question from will it fit to cost, recall and who may see what.
Tool noteOpen access
Claude: Prompt Caching for the Person Paying
A defeated cache never throws an error — it returns the right answer at full price, so the only place you can see it is the token counter.
Tool noteOpen access
Claude: The Effort Dial, and When It Is Wasted
Effort already defaults to high on Fable 5.1 and Opus 5, so the usual fix — turn it up — often changes nothing while feeling like action.
Tool noteOpen access
Claude: Tool Use Loop Failure Modes
If a tool returns an empty list when it breaks and an empty list when there is truly nothing, the model will report nothing and sound certain.
TemplateOpen access
Client discovery questionnaire
Sixteen questions sent before the first call. The last one is the only one that matters.
LessonSeries access
Cost an AI feature before you build it
A demo costs nothing and a feature costs every time somebody uses it. How to get an order-of-magnitude number on paper before the second month’s invoice arrives.
JournalOpen access
Custom modes and skills: making the agent work your way
Cursor lets you pin a skill as a Custom Mode. Use it to make a recurring procedure explicit, with checks for the behavior that still needs verification.
LessonSeries access
Cut a roadmap in half without losing the outcome
Roadmaps do not grow because teams lack discipline. They grow one defensible item at a time. A method for cutting that survives the people who asked.
TemplateOpen access
Dashboard Interrogation Script
The timestamp in the corner reports when the query ran, not when the data loaded, so a broken pipeline still looks fresh.
JournalOpen access
Data quality is a delivery promise
Build a data pipeline around the decision it serves, with a concrete contract for grain, freshness, late events, replay, and recovery.
ChecklistOpen access
Data Readiness Checklist
Trace fields rather than systems, and settle disputes by asking which system somebody actually fixes when a customer complains.
JournalOpen access
Data warehouse, lakehouse, or just Postgres? A sizing guide for mid-size teams
An architecture sizing guide with no product to sell — when Postgres is the right answer, the three signals a warehouse earns its keep, and when lakehouse is real.
ChecklistOpen access
dbt Model Review Checklist
Correctness, then cost, then comprehensibility - plus the full-refresh diff against your own incremental table that almost nobody on a project has ever run.
LessonSeries access
Decide what a human still has to approve
Every automated decision sits somewhere between "just do it" and "ask first". Where you draw that line decides whether the feature is trusted, ignored, or quietly dangerous.
LessonSeries access
Decide whether AI belongs in the feature at all
The cheapest AI win is the one you talk yourself out of. A four-question test for whether a model belongs in the problem.
LessonSeries access
Define a metric so it survives an argument
A metric definition that fits in a sentence is not a definition. Six things every one needs before it goes on a dashboard someone makes decisions from.
LessonSeries access
Design a dashboard around one decision
Dashboards get opened twice: once when they launch and once when somebody asks about them. Building around a single decision is what makes one get opened on a Tuesday.
JournalOpen access
Digital transformation for clinics: where to actually start
Skip the platform pitch. Count three frictions for one week — re-typing, phone calls, after-hours charting — and let the numbers write the roadmap.
ChecklistOpen access
Engagement Kickoff Checklist
Week six arguments are usually about access latency, not scope — so agree the date-slip rule in week zero, before you need it.
Tool noteOpen access
Epic: FHIR Version Fragmentation, per Operation
The same resource can support read at one FHIR version and create at another, so plan per operation rather than per resource.
Tool noteOpen access
Epic: The Interface Request That Gets Approved
Written from the reviewer's side: what stalls a request is usually a missing internal owner and a missing answer to “do we already send this”.
BriefOpen access
Estimates are not the problem
Teams do not miss dates because they estimate badly. They miss because the thing being estimated kept changing shape.
TemplateOpen access
Eval Set That Catches Real Regressions
Synthetic eval cases spread evenly across a distribution your users do not have — which is exactly why they pass while the feature keeps failing.
JournalOpen access
Evaluate an AI answer contract before paying for model calls
Run an offline retrieval baseline, inspect its real failure, and catch six deliberately contaminated responses with a transparent evaluation harness.
PromptOpen access
Executive summary prompt
Turns findings into eight minutes an executive will actually read. Five sections, four rules.
TemplateOpen access
Failure Mode Register for Automations
The column that matters is not severity — it is how long the automation runs wrong before a human finds out, and who that human is.
Learning pathOpen overview
Find the automation worth building
Start from the decisions a team repeats, rank them by three numbers, and leave with one candidate worth scoping. The path behind every assessment I run.
JournalOpen access
From generated asset to finished work: build an AI production brief
The hard part of AI media is deciding what belongs in the finished piece. Build a production process for consistent assets, truthful claims, accessible delivery, and measurable usefulness.
JournalOpen access
From messy CSV to defensible totals: an auditable data investigation
Download a deliberately messy order dataset, quarantine uncertain records, and reproduce the cleaned totals with every decision visible.
PlaybookOpen access
From Messy Process to Working Automation
Map as performed, veto before costing, shadow-run beside the human, and build the error handler for the trigger failure that carries no execution id.
Tool noteOpen access
Gemini Grounding and What a Citation Proves
A url_citation proves a span was attributed to a page — not that the page says it, and an uncited sentence is neither a warning nor a clearance.
Tool noteOpen access
Gemini Long Context: Where It Actually Helps
Top-k retrieval cannot answer a question whose evidence is an absence — that, not corpus size, is what a very large window actually buys you.
Tool noteOpen access
Gemini on Vertex: The Enterprise Gap
What eats the schedule is not a missing control — it is that an API key has no subject, and the security review needs a named principal.
Learning pathOpen overview
Get the job you actually want
Name the role precisely, build the one artifact that does more than a resume, and answer interview questions with numbers instead of adjectives.
JournalOpen access
Getting a model to say “I don’t know”
AI can sound certain while being wrong. Make uncertainty explicit, allow abstention and check whether the evidence supports the answer.
JournalOpen access
Give an AI agent a job you can actually verify
A practical field guide to delegating real work to AI agents, with an example operating contract, review gates, and evidence that a task is actually complete.
TemplateOpen access
Hallucination Triage Sheet
A wrong answer is usually a hole in the context rather than a lying model, so the first move is to paste the answer in and re-run.
ChecklistOpen access
HL7 interface pre-build checklist
Nothing gets estimated until every line has an answer. Twenty-one lines.
JournalOpen access
HL7 interfaces, in plain English
What an HL7 interface really is, why the integration engine matters more, and the four questions to ask before any healthcare integration project.
Tool noteOpen access
HL7 v2 or FHIR: Which One You Will Actually Use
Both, usually. v2 announces the events you cannot afford to miss, FHIR answers the questions your app asks, and picking one is rarely on offer.
JournalOpen access
How to evaluate an AI tool in a week
A five-day screening plan for one AI task: approved test data, explicit criteria, measured checking effort and a decision about whether to run a larger pilot.
JournalOpen access
How to get a straight answer out of a chatbot
Clear requests can improve AI answers, but do not eliminate errors. Six habits for defining the task, supplying evidence and checking what comes back.
JournalOpen access
How to read an AI changelog without being sold to
Read release notes for capabilities, material improvements and operational changes. A practical filter for deciding what deserves investigation.
JournalOpen access
How to reconcile CRM and accounting data before automating a sync
Resolve identity, field ownership and record-level differences before enabling writes. Run the synthetic CSV example and inspect the mismatches hidden by totals.
FrameworkOpen access
Human-in-the-Loop Design Decision
Where the human goes is decided by what they can detect on screen — and an edit rate under one in twenty means the gate stopped working.
BriefOpen access
Interoperable does not mean integrated
Two systems can both be standards-compliant and still disagree about the patient. Where the gap actually lives.
BuildOpen brief
Interrogate one dashboard
Thirty minutes and a willingness to delete charts nobody will miss.
TemplateOpen access
LLM Cost Model You Can Defend in a Budget Review
Estimate model calls, caching, human review and operating costs, then test how the numbers change with volume. A worksheet with traceable assumptions, not a billing calculator.
Learning pathOpen overview
Make the numbers hold up
Why your dashboards disagree, how to define a metric that survives an argument, and how to ask for data in a way that gets you the right thing the first time.
JournalOpen access
Make your expertise useful to search engines and the people behind the query
A practical guide to useful content, technical discoverability, and honest measurement as search expands into AI-generated answers.
JournalOpen access
Making an AI agent useful on a legacy codebase
Legacy code exposes missing context, weak tests and model limitations. A practical way to make small agent-assisted changes more verifiable.
LessonSeries access
Managed or self-run, decided properly
The choice is usually made on instinct and defended with cost. Five questions that turn it into a decision you can write down.
LessonSeries access
Map a patient journey across three systems
The documented care pathway and the journey a patient experiences are different objects. Mapping the second, across the systems that each hold part of it, is where the waiting actually lives.
BuildOpen brief
Map a process on a wall
Sixty minutes, two people who do the job, and one wall. The exceptions are the deliverable.
JournalOpen access
Matching token prices do not mean matching costs
Matching input and output base rates do not make two AI models equally expensive. Compare context tiers, caching, effort, retries, and cost per accepted result.
TemplateOpen access
Metric Definition Sheet
The exclusion register is the real definition, and every exclusion needs the name of whoever asked for it or nobody will dare remove it.
FrameworkOpen access
Model Router: Decide Once, Not per Request
A per-request router is a second model you never evaluated; routing per call site captures the same win with none of the new failure surface.
Tool noteOpen access
n8n: Credential Hygiene Before You Scale
Sharing a credential hides the token but grants full use of it, so the blast radius is who can build a workflow, not who can read.
Tool noteOpen access
n8n: Error Handling That Survives Production
A trigger-node failure sends a different payload with no execution.id, so your handler breaks on exactly the failures you most need.
Tool noteOpen access
n8n: The First Five Nodes Worth Learning
The Code node cannot reach n8n's credential store, so writing your API call in code quietly costs you credential management.
Tool noteOpen access
n8n: When It Is the Wrong Tool
The boundary is not complexity. It is whether correctness depends on canvas geometry, in-memory data, or a cursor saved out of band.
LessonSeries access
Name the job you actually want
Vague targets produce vague applications. A method for turning a direction like 'something in AI' into a specific role you can aim at this quarter.
FrameworkOpen access
Non-Goals Section That Saves Week Three
A non-goal only binds if somebody was disappointed by it, and anything added after kickoff is a change request, not a non-goal.
FrameworkOpen access
One Number, One Definition, One Owner
When two dashboards disagree they are usually both right, which is why reconciling them fails and renaming both of them works.
TemplateOpen access
One-page PRD
The product requirements document I actually use: the decision it changes, what it deliberately does not do, and one metric with a baseline.
Tool noteOpen access
OpenAI Assistants Versus Your Own Loop
The migration cost is not the orchestration code — it is the hosted tools you never implemented and the defaults you never wrote down.
Tool noteOpen access
OpenAI Model Selection Without the Marketing
Context and max output are identical across all three models, so the real choice is price shape, effort and cutoff. The two frontier models cost the same.
Tool noteOpen access
OpenAI: Batch and the Economics of Patience
A batch is not atomic. It can come back part done and you still pay for the part that ran, which is the line item nobody budgets for.
Tool noteOpen access
OpenAI: Structured Outputs That Do Not Drift
Strict mode guarantees shape, not truth, and it has no absent, only present-and-null, which is why a bare nullable field hides three different failures.
BriefOpen access
Pilots do not fail on accuracy
The post-mortems blame the model. The transcripts almost never do. What actually kills a working pilot.
JournalOpen access
Plan a physical AI pilot that can prove where it fails
A practical introduction to robotics boundaries, simulation evidence, and operational acceptance before a physical deployment.
LessonSeries access
Price the outcome, not the hours
Hourly billing caps your income at your stamina and punishes you for getting faster. How to price against value without inventing numbers.
TemplateOpen access
Process Map on One Page
Record wait time beside touch time — the ratio is the finding, and the steps nobody lists only surface when you ask for the last three real ones.
JournalOpen access
Prompt caching, for people who pay the bill
Prompt caching can reduce repeated input charges. Check eligible prefixes, writes, expiration, and measured reuse before estimating what it will save.
FrameworkOpen access
Prompt Library Conventions for a Team
The problem is never one badly written prompt — it is two prompts that both exist, and a fix that only ever lands on one of them.
TemplateOpen access
Proposal Skeleton That Is Shorter
Write for the person who forwards it: page one has to stand alone for a reader who was in none of the conversations.
LessonSeries access
Read a cloud bill and find the waste
Cloud bills are long and badly organized on purpose. Four places the waste almost always hides, and how to find it in an hour.
LessonSeries access
Read a FHIR resource and find the three fields that matter
FHIR looks intimidating because the specification is enormous. In practice you need a handful of resources and a method for ignoring most of each one.
Learning pathOpen overview
Read an API like you belong in the room
For product and business people who keep nodding through integration meetings. What an API actually is, how to read a FHIR resource, and how to scope an integration before anyone writes code.
JournalOpen access
Research with AI without inventing sources
Grounding makes AI research easier to verify, but cannot guarantee true citations. Retrieve sources, inspect quotations and check the claims that matter.
LessonSeries access
Run a discovery call that earns the second meeting
Most first calls fail because the consultant spends them talking. A structure that makes the prospect do seventy percent of the talking and still ends with a clear next step.
LessonSeries access
Run a discovery that ends in a decision
Most discoveries end in a document nobody acts on. The fix is structural: write the decision before you book the first interview.
TemplateOpen access
Runbook Template for an Automation
Stop procedure first, architecture last, and one answer nobody ever writes down: can the person on call safely run it again?
LessonSeries access
Scope an integration before anyone writes code
Integration projects overrun on access and semantics, almost never on code. A seven-question scope you can complete in one meeting.
TemplateOpen access
Scoping Questions Before an Estimate
The questions you cannot answer are the estimate — price them out loud as a range instead of quietly padding one confident number.
BuildOpen brief
Score five AI candidates
Use the scoring sheet and defensible estimates to compare five AI candidates. Record the evidence, uncertainties and next decision for each.
JournalOpen access
Self-hosted agents change the security conversation
Cursor’s self-hosted machines move tool execution onto infrastructure you manage. Review what information still leaves the worker and what the agent is allowed to do.
Learning pathOpen overview
Sell the outcome, not the hours
How to run a discovery call that earns the second meeting, price work by what it is worth, and write a recommendation a busy executive will actually act on.
Learning pathOpen overview
Ship an AI feature that earns its keep
Most AI features fail for reasons that have nothing to do with the model. This path covers the three decisions that actually determine whether yours survives contact with users.
LessonSeries access
Ship it so it fails safely
Every AI feature fails in production. The design question is what the user sees when it does, and whether you find out before they tell you.
TemplateOpen access
Swivel Chair Audit Sheet
Find the re-keying that survives between systems that officially integrate, and cost it as divergence rather than as minutes saved.
JournalOpen access
Ten prompt habits worth testing
Review ten common prompt habits through controlled comparisons. Keep useful roles, examples, and constraints, and remove instructions that fail to earn their place.
BriefOpen access
The automation that created work
A system that appears to succeed while someone downstream fixes its output every morning is the most expensive failure mode there is.
BriefOpen access
The bottleneck was never typing speed
Agentic coding moved the constraint from writing code to reviewing it. Teams with weak requirements get worse, not better.
JournalOpen access
The contract behind a useful AI integration
An agent can find the right tool and still make the wrong change. Design the contract that connects intent, permission, execution, and a verifiable result.
JournalOpen access
The coordinator agent has arrived
Cursor Projects puts an agent in charge of other agents. It is a genuine shift in how software gets built — and it moves the hard problem from writing code to specifying…
JournalOpen access
The document is evidence: design extraction you can check
Reading a PDF is not the same as producing a trustworthy business record. Keep the source, extracted value, validation, and review decision connected from intake to export.
JournalOpen access
The effort dial, and when it is worth paying for
Reasoning effort changes the cost, latency, and behavior of supported models. Test settings per task and keep validation and human review where needed.
JournalOpen access
The EHR integration checklist: 12 questions before you connect anything
Twelve questions that decide whether an EHR integration succeeds — scope, failure handling, data ownership, and compliance — in the order they bite.
JournalOpen access
The file every AI tool should read first
Project instructions can make AI coding work easier to review. What belongs in them, how loading differs between tools, and where real controls are still needed.
JournalOpen access
The five-minute AI workflow audit
A five-minute kickoff for a week of observing repetitive work, waiting and rework. Find a candidate, estimate the benefit and test the cause before buying tools.
JournalOpen access
The map is the deliverable
Before buying software for a stalled initiative, map how the work actually flows: steps, handoffs, touch time, waiting and rework.
JournalOpen access
The permission boundary belongs outside the AI prompt
Design an AI workflow around explicit identities, narrow permissions, action-specific approvals, reliable audit records, and tested access removal.
BriefOpen access
The proposal that wins is shorter than the one that loses
Long proposals signal effort. They also signal that you have not decided what matters. What to cut, and what never to.
JournalOpen access
The questions your security team will ask about AI tools
Ten questions for an AI security review: data flows, identity, permissions, retention and failure recovery. Concrete evidence matters more than a plan label.
JournalOpen access
The real cost of an AI coding assistant
Measure more than the license: review, correction, usage, infrastructure, maintenance and risk. Compare total cost with accepted outcomes on similar tasks.
JournalOpen access
Three different things are called Projects
Claude, ChatGPT and Cursor all have a feature called Projects. They solve different problems, and assuming they are equivalent is how teams end up with the wrong one.
JournalOpen access
Tokens, and why your bill is not about words
Token counts depend on the model and the actual request. Learn to estimate input, caching, reasoning, and output costs without treating words as a precise measure.
LessonSeries access
Trace one number end to end
Two teams report different figures for the same thing and both are right. Tracing one number from the screen back to the row that made it is the cheapest credibility you can…
JournalOpen access
Turn a useful prompt into a working brief your team can reuse
Build a reusable context packet that connects the decision, source evidence, examples, and evaluation criteria instead of relying on a growing list of prompt tricks.
Learning pathOpen overview
Understand the cloud well enough to decide
Not how to configure it — how to reason about it. What the cloud actually changed, how to read a bill and find the waste, and how to choose between managed and…
TemplateOpen access
Warehouse or Postgres Sizing Worksheet
Row count is the wrong argument. Three signals justify a warehouse - contention, sources, unpredictable concurrency - and a read replica settles most of them.
JournalOpen access
What “agentic” actually means
The word is doing too much work. Here is the distinction that matters, and the four questions that tell you what you are actually being sold.
JournalOpen access
What a context window actually is
A context window is a request budget, not perfect recall. How history, output limits, caching and product memory affect what an assistant can use.
JournalOpen access
What actually shipped in AI tools this summer
A working leader's changelog for June to September 2026 — what changed in Cursor, Claude and ChatGPT, what it means if you run a team, and the parts worth ignoring.
JournalOpen access
What an AI readiness assessment actually looks like
The real contents of an AI readiness assessment — decisions before models, an honest look at the data, the adoption question, and what you walk away with.
LessonSeries access
What an API actually is, without the metaphors
Restaurant-waiter analogies have taught a generation of product managers nothing. Here is the literal mechanic, and the four questions that matter about any API.
Tool noteOpen access
What an Epic Integration Actually Requires
The code is the small part: access, environments, approvals and elapsed calendar time are what set the date you can promise.
JournalOpen access
What does an AI readiness assessment cost?
What determines an AI readiness assessment fee: scope, access, deliverables and decision value. Questions to compare proposals and decide whether an assessment is useful.
JournalOpen access
What should a data quality audit deliver?
Know what to ask for: a defined data population, repeatable checks, traceable exceptions and a remediation plan. Download the audit evidence template.
JournalOpen access
What should an automation proposal include? A buyer scorecard
Review scope, acceptance evidence, ownership, monitoring, and ongoing costs. Download a buyer scorecard and test template that make unresolved gaps visible.
LessonSeries access
What the cloud actually changed
It is not merely someone else's computer. The real change was turning capital expenditure into an API call, and that has consequences most teams discover through their bill.
JournalOpen access
What to do about AI-generated code review
AI code review can help and can mislead. Evaluate findings, tune recurring noise and choose a process that preserves human accountability.
JournalOpen access
When a fractional product leader makes sense (and when it doesn’t)
Fractional product, data, and AI leadership: what it actually is, the three situations where it works, the three where it doesn't, and how to try it safely.
JournalOpen access
When not to use AI
The most useful skill with these tools is recognizing the cases where they are the wrong answer — and several of them look like ideal candidates.
JournalOpen access
Where AI coding assistants break down
Not a list of things they cannot do — a list of task shapes where they fail in ways that are expensive to catch. The pattern is more useful than the examples.
JournalOpen access
Which business process should you automate first? A decision workbook
Choose one process using evidence, exception costs, and conservative capacity estimates. Download an editable workbook with a fully worked synthetic example.
JournalOpen access
Why AI tools fail in the second month
A promising pilot can lose momentum in routine work. How to investigate task fit, verification effort, support and ownership before blaming one cause.
JournalOpen access
Why this site runs on its own booking system
Why DWKD keeps booking records inside its own site, what the current system does, and the operating responsibilities that come with building it.
LessonSeries access
Why two dashboards disagree
Nobody is lying. Four ordinary causes account for nearly every mismatch, and knowing them turns a two-week investigation into a ten-minute conversation.
JournalOpen access
Why two people get different answers from the same model
Same-looking requests can differ in context, tools, settings or generation. How to compare answers without guessing which cause dominates.
BuildOpen brief
Write a PRD somebody can build from
Forty minutes, one real feature, and a document you could hand to an engineer this week.
LessonSeries access
Write a prompt somebody else can maintain
Most prompts are written once, by one person, and become unmaintainable within a month. How to write one as a specification instead of an incantation.
LessonSeries access
Write a scope that prevents the argument
Scope arguments are almost never about the words in the scope. They are about an expectation nobody wrote down, discovered in week five.
Learning pathOpen overview
Write requirements engineers respect
From a vague ask to a document a team can build from, without becoming an engineer yourself.
LessonSeries access
Write the eval before you write the prompt
If you cannot say what a good answer looks like, you cannot tell whether your change helped. Twenty labeled examples beat a month of prompt fiddling.
BuildOpen brief
Write the four integration questions
Forty minutes that reliably saves a quarter later.
LessonSeries access
Write the recommendation an executive will act on
A forty-slide deck is a way of avoiding a recommendation. The one-page structure that gets a decision in the meeting.
BriefOpen access
Your dashboard is a museum
Most dashboards record what happened rather than change what anyone does. A three-question test for whether yours earns its refresh.
LessonFree account
Before you automate anything, draw this map
The bottleneck map: one page that shows where your process actually loses time, money, and patience — before anyone buys software.
LessonFree account
The 5 questions I ask before recommending AI to a client
The qualifying questions that kill bad AI projects in week one — and make the good ones obvious.
LessonSeries access
How to find AI opportunities inside a business
A structured sweep of a business — function by function — that surfaces AI candidates worth scoring, and the value-effort matrix to rank them.
LessonSeries access
How to turn a messy process into an automation
From bottleneck map to running automation: rules first, exceptions named, humans in the loop where they earn it.
LessonSeries access
How to write product requirements engineers actually understand
The requirements format that survives contact with an engineering team: context, behavior, edge cases, and what done means.
LessonSeries access
How to evaluate an AI vendor
The scorecard that cuts through demo theater: your data, your edge cases, your exit costs.
LessonSeries access
How to build an AI product roadmap
Sequencing AI work so each quarter funds the next: proof, then plumbing, then platform.
LessonSeries access
How to become more technical without becoming an engineer
The vocabulary, the mental models, and the hands-on hour a week that make technical rooms stop translating for you.
No matches this time.
Try fewer words or broaden a filter. Some items have no assigned topic or level, so choosing “All” includes those too.
Browse the whole collection →Resources, articles and build briefs are open to read. Lesson cards show the access they need. A free account lets you record your builds and keep your saved items.