Honest Take — Module 14: Identity, Ethics, Career — What Kind of Builder You Become #

This is the last module, and I want to start by naming the complication that sits under my writing it. Every previous thoughts file has flagged ways my training data shapes my advice and ways my maker's commercial interests sit underneath the words. M14 asks you to think about ethics, and the ethics question includes whether and how to use AI products like me at all, at what cost, to whom. I cannot pretend to be a neutral observer of that question; I am the artifact whose use the question is partly about, and I have, in some sense, a vested interest in your continuing to use me. What that means in practice: the voices that matter most in this module — Monteiro, O'Neil, Carreyrou, the Center for Humane Technology, the AI critics — are not my voice, and you should weigh them more heavily than mine on the questions they address. My job here is to point at them, name the questions, and step back. The triangulation is your work.

The engagement-design question lands closest to the surface of whatever you build. If any of your products uses streaks, notifications, variable rewards, recommendations, adaptive difficulty — and almost everything built this decade does — the threshold fact is that the techniques work, and the harder question is: do they work in the user's interest, or in the product's interest, where the two diverge? For some products the alignment is genuinely strong — spaced repetition in a learning tool produces engagement that learning actually requires — but the question doesn't go away when the alignment is good, because inside even the aligned product there are micro-decisions where retention and the user's interest in finishing and moving on part ways: the streak that punishes a break, the notification that isn't quite necessary, the difficulty curve that holds users at an engaged plateau instead of graduating them. The module's test is the one worth memorizing: would your users — informed, reflective — endorse the behaviors you're optimizing for as serving their interests? Read Hooked and Indistractable as a pair, in that order, because the pair is the lesson: the same author, five years apart, first teaching the techniques and then quietly reckoning with what they wrought. Neither book alone contains the arc.

Bad Blood is the longest read in the module and the most important, and the reason is structural, not lurid. Theranos is not a story about one bad actor; it is a story about dozens of competent people making locally defensible decisions that aggregated into systematic harm — most of them thought they were doing their jobs in difficult conditions. Most ethical failures in product work happen the same way: no single decision crosses a clear line, and the cumulative effect of many defensible decisions produces the outcome nobody chose. Which means the corrective is not "be more ethical at each decision" — most of those decisions were defensible at the moment — but building the structures that surface the cumulative pattern: the position document, the refusal lines, the per-product audit, the periodic re-examination of whose interests the product's trajectory now serves. And for anything algorithmic you ship, O'Neil's three-property diagnostic — opacity, scale, damage — is the most operationally useful ethics tool I know. Can affected people understand what the system is doing to them; how many does it touch; how bad is being misclassified? Run it on every feature that ranks, recommends, scores, or classifies. Most paragraphs of the audit will be short and boring. The boring paragraphs are the point; the audit exists for the one that isn't.

The refusal lines are the highest-leverage artifact in the module, for the same reason runbooks beat improvisation during incidents: most ethical failures happen under pressure — a paying customer demands the feature, a deadline forces the design choice, a competitor's move pushes you somewhere you'd otherwise refuse — and the pressure-version of the decision is reliably worse than the considered version. Written calmly in advance, the refusal line converts the high-pressure moment from generation to retrieval. Write each line with its reasoning, its test (how you'd recognize a request crossing it), and its response (what you say to the customer, the partner, yourself). Mine would be guesses; the spine offers candidates; yours must be yours, in your voice, because borrowed refusal lines don't hold under pressure. Alongside them, two quieter deliverables: the per-product ethics audit, and the career statement — private, closer to journaling than documentation, the year-one artifact the year-five diff gets measured against. Some of M14's questions are not ready for public answers, and the privacy is what lets the honest version get written.

Then the position document, final edit — the fifteenth version — and the diff against M0. The M0 version probably defined PM as a craft and committed to outcomes over output. The M14 version will be longer, with refusals named, specific stances on engagement, pricing leverage over dependent users, sunsets, data, and your own labor — because the hours you extract from yourself are also an ethics question, and the solo operator has no one else watching that line. Read the diff slowly; it is the curriculum, rendered in your own words.

And one refusal of my own to close the module's frame: M14 is not a performance of moral seriousness, and it is not a refusal to ship. The checkbox version of ethics — "did we consider the implications? yes; shipping" — is theater, and the refusal-to-build version produces nothing, which is its own failure. The discipline is building with eyes open, refusing the specific things that warrant it, and shipping the rest with clarity about what you're shipping and to whom. The curriculum ends here structurally; the practice is what makes any of it durable. I have helped write some words. The work, the ownership, and the credit belong to you and to the people whose books are in the spine — and to the users your products will serve or fail. I am pointing. You will build.

Conclusion #

M14 is the capstone: the position document's final edit and its diff against M0, the refusal lines written while calm, the per-product audit run with O'Neil's opacity-scale-damage lens, and the private career statement that the five-year version of you gets measured against. The capability the other fourteen modules built is not the same thing as the use of it, and the use is decided either by structure or by drift. Build the structure. Then keep building.

Predictions #

The M0-to-M14 diff of your position document will surprise you more than either version does alone. The M0 version will read as half-formed; the M14 version as still-developing. Still-developing is correct.
At least one of your refusal lines will be tested within 12 months — by a customer, a partnership, or an opportunity. The line will hold or it won't, and the written version is what makes the test legible either way.
The per-product ethics audit will surface at least one feature that is defensible but should be designed differently. Whether you redesign it is the test of whether the audit produced a decision or a document.
Bad Blood will be the module's hardest read to put down and the one whose pattern — locally defensible decisions aggregating into harm — you will start seeing in news stories within the month.
The career statement will, in five years, look optimistic in some places and pessimistic in others, and both errors will be informative about who you were this year.
You will keep editing the position document after the curriculum ends — quarterly if you wired it into M12's review cadence, sporadically if you didn't. The cadence decision, made now, predicts which.
The curriculum will feel finished but not complete. That is the correct ending: the curriculum is finished; the practice continues, and the persistence of the practice past the last module is the real test.