Podcast: How Will We Train Developers If AI Does the Routine Work: A Conversation with Scott Hanselman

Michael Stiefel (host) with Scott Hanselman (guest, VP, Microsoft/GitHub) · podcast (InfoQ Architects Podcast; Reader category: rss) · 2026-09-14 Reader 🖥️

practical · mixed — Primarily practical/argumentative: the conversation argues the industry ought to adopt a specific organizational mechanism (the preceptor role). It is carried substantially by narrative (the god-object anecdote, the nursing analogy, the wax-on-wax-off metaphor, the seed-corn analogy) rather than abstract argument, and closes with an expository personal-reflection segment (the Architect's Questionnaire) that is mostly color rather than thesis-bearing.

Structure

Key terms

• preceptor / preceptorship model
  An explicit, formally designated senior-engineer role -- borrowed directly from nursing's teaching-nurse system -- whose job performance is evaluated on how well they train and produce new senior engineers, not on how much code they personally ship or how much they resemble a manager or tech lead.
• wax on, wax off
  The Karate Kid metaphor for the tedious, repetitive, low-stakes toil junior engineers were traditionally assigned; Hanselman's point is that this toil was not incidental busywork but the actual mechanism through which craft skill was absorbed.
• engineering thrive / time to first PR
  A deliberately non-productivity metric tracking how a new hire feels about landing their first pull request, used as a signal of personal growth and enablement rather than output.
• the third place
  The informal social space distinct from work and home; Hanselman's claim is that remote work, social media, and AI collapse work, home, and the third place into one undifferentiated place, eliminating a developmental resource.
• architecture (vs. features)
  The big-picture, systemic design judgment distinct from generating working feature code; Hanselman's specific claim is that AI is 'myopic' and characteristically bad at this, evidenced by AI accumulating an undisciplined 'god object' in his own current project.
• too agile
  Hanselman's claim that AI has not killed agile but has taken its fast inner-loop/outer-loop philosophy to an unstable extreme, where speed without discipline causes the work to 'wobble' and risk flying off the road.

Propositions

• explicitAI generates software features well but performs poorly at software architecture -- it is myopic and misses the big picture, as evidenced by AI accumulating a 'god object' in Hanselman's own current project that he now has to remove.conf: high
  supportsShould You Still Become a Software Engineer in 2026? GitHub VP · p10 — Same claim (AI generates features well but is architecturally myopic, producing 'god objects') corroborated with a fresh, concurrent anecdote from Hanselman's own current project, in a different venue.
• explicitcorroboratedThe industry should adopt a formal 'preceptor' role, borrowed from nursing, whose explicit job is to train new engineers and whose own performance is evaluated on how well they mint new seniors, not on how much code they ship.conf: high
  extendsShould You Still Become a Software Engineer in 2026? GitHub VP · p5 — Same preceptor proposal, elaborated here with the full nursing analogy, explicit evaluation criteria, and a completed-first-cohort pilot report.
  supports ←The Bet On Juniors Just Got Better · p2 — Both put learning/mentoring outcomes ahead of raw output as the thing to manage and evaluate for -- Hanselman at the level of the preceptor's own performance review, Beck at the level of how a junior's day-to-day work should be directed.
  extends ←Redefining the Software Engineering Profession for AI · p6 — This ACM article is the formal, co-authored (with Russinovich) elaboration of the same 'preceptor' proposal Hanselman floated informally -- adding concrete 3:1-5:1 ratios, an 'EiC mode' coding-assistant design, and chat-log review, though without restating the earlier piece's sharper point that preceptors must be evaluated on mentoring outcomes rather than their own throughput.
• explicitCalling new hires 'interns' or 'apprentices' creates a status asymmetry ('upon entrance, I'm less than') that makes entry-level people expendable and interchangeable rather than people the organization is accountable for growing.conf: high
• implicitdevelopedThe routine, tedious work traditionally assigned to junior engineers ('wax on, wax off') was not incidental busywork but the actual mechanism by which craft skill was transmitted -- remove it and there is no obvious substitute path by which novices absorb expertise.conf: high
  supportsWhen AI can do more than we can check · p5 — Both argue that automating junior-level routine work removes the practice-based mechanism through which novices historically absorbed expertise.
  supportsResearch: Why Some Junior Employees Work Well with AI—and Others Don’t · p8 — Corroborates the claim that AI is displacing the traditional repetition-based entry-level on-ramp for professional judgment.
  extends ←LIVE: Uncle Bob on Software Fundamentals in the Age of AI · p17 — Hanselman's claim that juniors' routine tedious work was the actual mechanism of craft transmission, now threatened by AI absorbing that work, is extended here with a concrete proposed substitute: deliberately make the junior do agent-style constrained, deterministic-gated tasks instead of the old routine work, to recreate the same transmission mechanism in the new regime.
• explicitRemote work does not help early-career people grow; in-person interaction (whiteboards, handshakes, physical presence with senior engineers) provides developmental value that virtual interaction does not replicate.conf: high
• explicitThe convergence of remote work, social media, and AI collapses work, home, and 'the third place' into a single undifferentiated place, which prevents people from growing into fully formed, healthy individuals.conf: high
• explicit'Engineering thrive' (time-to-first-PR) should be tracked and treated as a morale/personal-growth metric, not a productivity metric.conf: high
• explicitPreceptorship is easier in smaller organizations (Hanselman's 400-500 person 'sweet spot') but is achievable at scale in large companies if implemented in smaller shards with explicit executive buy-in and support.conf: high
• explicitInvesting in preceptorship costs money and slows output in the short term, but not doing it is far more expensive later, when companies are forced to poach seniors from competitors who then import a different (mismatched) culture and technique.conf: high
  extendsEngineering Leadership Report 2026 · p18 — Supplies a mechanism (poached seniors import competitor culture/technique) explaining why the leadership-reported junior-pipeline execution gap persists despite being widely named as a concern.
• explicitThe deeper cause of the junior-pipeline collapse is not AI itself but pre-existing short-term 'hyperoptimization' and quarter-by-quarter corporate thinking; 'AI is not going to take your job -- the hyper optimization of some C-suite executive is going to take your job.'conf: high
• explicitAnthropomorphizing AI assistants (giving them names and voices, treating them as companions) is unhealthy, and the risk will worsen as embodied/robotic AI becomes more expressive and lifelike.conf: high
• explicitEven while heavily using AI-generated code, human engineers must retain final review and coordination authority -- 'human beings have got to keep their hands on the wheel' -- checking for architectural soundness and AI 'slop' before merging.conf: high
• explicitAI has not killed agile; it has pushed agile's fast-feedback-loop philosophy to an unstable extreme ('too agile') where speed without discipline risks the work 'wobbling' and flying off the road.conf: high
  supportsNo, AI Did Not Kill Agile · p1 — Both conclude AI has not literally killed agile, though Hanselman frames the distortion as agile pushed to an unstable speed extreme rather than a speed/knowledge conflation.
• explicitUniversity coursework (small group projects) does not teach the real software development lifecycle; that gap increasingly has to be closed by industry-embedded internships and cohorts, ideally consuming most of the senior year.conf: high
• implicitUsing AI to eliminate junior-level training work is a form of consuming a renewable resource (the next generation of senior engineers) as a one-time input -- 'you're a farmer and you're eating your seed corn instead of planting it for the next year.'conf: high
• explicitAI productivity gains should be partly converted into reduced work rather than absorbed entirely as increased output, because open-ended, ever-compounding growth within a finite system is unsustainable ('that's cancer').conf: high
• explicitAI-mediated work correlates with loneliness and a loss of 'flow state' -- a psychological cost of AI adoption that is under-acknowledged relative to its productivity benefits.conf: high

Arguments

Narratives

• The 'god object' anecdote: over several weeks, AI assistance on Hanselman's current project accreted an undisciplined super-struct he now has to remove before it 'collapses under its own weight.'
  → Makes the abstract claim 'AI performs poorly at architecture' concrete and personally credible -- he is not speaking hypothetically, he is dealing with it as he records the podcast.
• The nursing-preceptor analogy: Hanselman's wife is a nurse preceptor, and nursing has a formal, metriced teaching-nurse role distinct from general nursing.
  → Supplies the entire structural template for the central policy proposal, letting Hanselman carry the argument by borrowed institutional design rather than abstract organizational theory.
• 'Wax on, wax off' (Karate Kid).
  → Reframes tedious junior work as pedagogically load-bearing rather than wasteful, making the loss of that work feel like a real deprivation rather than a pure productivity gain.
• The seed-corn farmer analogy, offered by Stiefel and enthusiastically endorsed by Hanselman ('the analogies are legion').
  → Reinforces the intergenerational-capital framing: consuming today's training material forecloses tomorrow's harvest of senior engineers.
• Hanselman drives to Seattle every four to six weeks 'to make sure that my badge still works and that the young people are still present,' despite being fully remote his entire Microsoft career.
  → Demonstrates he personally practices the anti-isolation prescription he argues for, giving the claim about in-person value personal credibility rather than being purely prescriptive.
• Personal maker anecdotes: the 3D-printed garage-door button cover, the DIY Peloton alternative ('PedalScape'), and the 30-year-old Visual Basic app still running at Pioneer Courthouse Square.
  → Establishes Hanselman's long-term maker/teacher identity and his valuing of software that 'holds up under pressure' for decades -- retroactively explains why he cares so much about training successors rather than just shipping features.
• The Y2K joke: 'they're going to go and pull me out of retirement in 2035 to go and fix the next big date bug.'
  → Dark-humor illustration of the concrete cost of pipeline collapse -- an industry forced to summon retirees because it failed to grow successors, foreshadowing a repeat of a known historical failure mode.

Questions addressed

• ✅ How will the next generation of software engineers be trained if AI absorbs the work juniors used to learn from?
  Adopt a formal preceptor role, evaluated on mentee growth rather than shipped code, modeled on nursing.
• ✅ What should an early-career engineer's on-ramp look like when 'hitting the ground running' is not realistic?
  Track 'engineering thrive' (time-to-first-PR) as a morale/growth metric, and insert deliberate, conscious growth moments at every step rather than assuming AI-mediated speed substitutes for development.
• ✅ Does remote work help or hurt mentorship?
  Hurts, in Hanselman's view -- though he says it can be 'pushed through' with deliberate metacognitive effort and periodic in-person contact.
• ✅ Has AI killed agile?
  No -- but it has driven agile's fast-loop philosophy to an unstable extreme ('too agile') that requires deliberate discipline (Kanban, user stories, a real SDLC) to keep from destabilizing.
• ✅ Is AI the cause of the documented decline in entry-level hiring since 2017?
  No -- Hanselman attributes it mainly to pre-existing corporate short-termism and 'hyper optimization,' with AI cast as an accelerant or scapegoat rather than the root cause.
• ⭕ What are the psychological/sociological costs of AI-mediated work?
  Named (loneliness, loss of flow state, anthropomorphization risk, the collapsed 'third place') but not systematically addressed -- the only concrete remedies offered are the preceptor program and Hanselman's personal practice of periodic in-person travel.

Gaps

• illogicalHanselman argues both that AI is centrally responsible for eliminating the junior-training on-ramp (the entire premise motivating the preceptor proposal) and that 'AI is not going to take your job -- the hyperoptimization of some C-suite executive is going to take your job,' attributing the pipeline collapse mainly to short-termism rather than AI. These two framings are never reconciled: is AI the proximate mechanism, or merely the current instrument of a short-termism that predates it? The conversation never specifies what a hyperoptimizing company without AI would have done to the junior pipeline, so the counterfactual attribution is left unexamined.
  Read next: Look for empirical work separating pre-2017 entry-level hiring decline trends from the post-ChatGPT AI-coding-adoption period, to test whether AI is additive to or merely relabeling a pre-existing trend.
• incompleteThe preceptor proposal is evaluated only on the promise that preceptors will be 'metriced' on 'how well they train people,' but no metric is actually specified beyond the anecdotal claim that Microsoft's 'first cohort of preceptors... has been very successful' and the qualitative 'engineering thrive'/time-to-first-PR framing, which is itself explicitly described as a morale signal, not a competency signal. There is no description of what a preceptor's performance review actually measures when it comes time to say a trainer 'shouldn't be a trainer.'
  Read next: extract-hbr-mentoring-matters-ai-era's more specified model (structured cadence, explicit competency goals, a written one-page agreement) offers a concrete alternative worth comparing against Hanselman's still-informal 'it's been very successful.'
• incompleteHanselman never addresses whether the preceptorship model as described (pairing, fellowship, code review) actually transmits architectural judgment -- the very skill he says AI cannot do -- or only reproduces the narrower skills the old toil-based system produced. If AI is already doing entry-level toil, and preceptors are pairing/reviewing rather than assigning toil, it is unclear preceptors reproduce the 'wax on, wax off' mechanism he says was essential; this is a gap between his own diagnosis (what created seniors before) and his own prescription (what he proposes now), which may be a different learning mechanism entirely.
  Read next: Look for accounts of what specifically preceptor/pairing sessions consist of at Microsoft's pilot -- do trainees do hands-on toil under supervision, or primarily observe and discuss AI-generated diffs?
• incompleteNo engagement with organizations that lack Microsoft's scale, cash, or executive sponsorship -- a resource-constrained startup, a non-tech company's small engineering team, or a highly outsourced/offshore engineering org has no obvious path to implementing preceptorship as described (400-500 person 'shards,' formal cohorts, explicit executive buy-in).
• incompleteThe 'third place' collapse and anthropomorphization-of-AI risks are raised as real sociological costs, but Hanselman offers no organizational remedy beyond his own personal practice (periodic travel to Seattle); there is no discussion of what fully distributed, no-office companies could do instead, nor any engagement with whether his prescription (in-person contact) is feasible for organizations that chose remote-first specifically to access geographically dispersed talent.

Closes gaps elsewhere

• The Hidden Realities of AI Adoption · g4 — That gap notes neither speaker proposed a candidate organizational design for the junior-talent pipeline problem beyond 'human above the loop.' This document proposes a concrete one: a formally named, nursing-style preceptor role, evaluated on trainee growth rather than shipped code, with company-size implementation guidance.
• Should You Still Become a Software Engineer in 2026? GitHub VP · g1 — That gap notes no mechanism was given for why individual-firm preceptorship doesn't just subsidize poachers. Here Hanselman partially addresses it: poached seniors 'use their competitor's techniques and their competitor's culture,' giving firms a private-benefit reason to grow their own people rather than rely on poaching -- though this argues training is still worthwhile despite leakage, not that the free-rider/collective-action problem itself is solved.
• We've Stopped Arguing About Whether AI Changes Software · g3 — That gap notes apprentice/pairing/rotation models were 'floated' with no clean answer for growing judgment in an agent-first world. This document supplies a specific, named model -- preceptorship plus explicit pairing/fellowship -- with a concrete evaluation criterion (preceptors are metriced on producing new seniors, not on code shipped).

Related documents

• Engineering Leadership Report 2026
• No, AI Did Not Kill Agile
• Research: Why Some Junior Employees Work Well with AI—and Others Don’t
  AI Labor Disruption Learning Culture & Slack Systemic Apprenticeship & Skill Transmission
• When AI can do more than we can check
  AI Labor Disruption Human-in-the-Loop Systemic Apprenticeship & Skill Transmission
• Have your agent record video demos of its work with shot-scraper video
  Agentic Engineering Human-in-the-Loop
• Agentic Skill Decay
  Learning Culture & Slack Systemic Apprenticeship & Skill Transmission
• [FULL WORKSHOP] AI Coding For Real Engineers - Matt Pocock, AI Hero
  Agentic Engineering Human-in-the-Loop
• The "funhouse mirror": How AI reflects the hidden truths of your software pipeline
  AI Labor Disruption Human-in-the-Loop
• Arvind Narayanan on Why AI Isn't All That Revolutionary
  AI Labor Disruption Human-in-the-Loop
• The Bet On Juniors Just Got Better
  Agentic Engineering Systemic Apprenticeship & Skill Transmission