Shu-Ha-Ri, Briefly
Shu ("keep, obey") is the apprentice stage: follow the established form exactly, without deviation, until it is internalized. Ha ("detach, digress") is where the practitioner starts adapting the form to context, breaking small rules deliberately because they now understand why the rules exist. Ri ("leave, transcend") is mastery: the form is no longer consulted, the practitioner creates their own approach and can teach it to others.
The model is old, but it maps onto software and database work almost without translation. A junior DBA who runs the runbook exactly as written, who writes the query the way the style guide says, who asks before deviating, is in Shu. A senior engineer who knows when the standard index strategy is wrong for this specific workload is in Ha. The architect who invents a new failover pattern because none of the existing ones fit is in Ri.
The part everyone skips over: there is no shortcut from Shu to Ri. Nobody arrives at "I know when to break the rule" without first spending years finding out, often painfully, what the rule was protecting against.
We Already Ran This Experiment Once
Offshore outsourcing was the first large-scale attempt to answer the question "do we really need to pay for the apprentice stage locally?" The logic in the 2000s and 2010s was straightforward: routine development, support tickets, and maintenance work, the Shu-level tasks, could be done cheaper elsewhere. Keep the architects and the client-facing seniors in Germany, the UK, or the US. Send the rest offshore.
It worked, in the narrow sense that costs went down. It also produced a side effect that took a decade to become visible: the offshore teams doing the "junior" work did not stay junior. They went through their own Shu-Ha-Ri progression, on someone else's payroll, and a meaningful share of the world's actual senior talent now sits in Bangalore, Warsaw, and Manila, not because of any inherent advantage, but because that is where the apprenticeships were still happening. Meanwhile, the onshore organizations that stopped hiring juniors twenty years ago are now short on domestic seniors, because the pipeline that produces them was switched off.
AI Is the Same Move, Faster and Closer to Home
The current argument is a rerun with a new mechanism. AI coding assistants, query generators, and automated runbooks can now do a large share of what used to be junior work: writing the first draft of a script, generating boilerplate, drafting a query, summarizing a log file, even proposing a fix for a known error pattern. The organizational conclusion writes itself, and plenty of leadership teams are already saying it out loud: why hire and train a junior when a senior plus an AI assistant covers the same ground, faster, without the multi-year ramp-up?
This is a more dangerous version of the same mistake than offshoring was, for one specific reason: offshoring at least preserved the Shu stage somewhere, for someone. It moved the apprenticeship, but it didn't delete it. Replacing junior work with AI output doesn't relocate the Shu stage. It removes it. There is no team of juniors quietly becoming seniors on the other end of an AI model. There is just less low-stakes work available for a human being to learn on.
| Mechanism | What Happened to the Shu Stage | Where the Future Seniors Come From |
|---|---|---|
| Offshore outsourcing | Relocated, not removed | The offshore teams themselves, over time |
| AI-assisted delivery | Automated away | Unclear, possibly nowhere |
The Kitchen Runs the Same Experiment
A classical kitchen brigade is a Shu-Ha-Ri pipeline with knives. A commis chef, the kitchen's apprentice, works the recipe exactly as written, under direct supervision, for years, before being trusted to season without measuring: that's Shu. A chef de partie or sous chef has enough accumulated judgment to rescue a sauce that's breaking, or extend a cook time because today's fish is unusually thick: that's Ha. A head chef who builds a menu nobody has seen before, who invents a technique instead of executing one, is operating in Ri. Every serious chef came up through the first stage in someone else's kitchen, peeling vegetables and getting a dish sent back, long before being trusted with the tasting menu.
| Kitchen Role | Shu-Ha-Ri Stage | IT Equivalent |
|---|---|---|
| Commis chef (apprentice) | Shu | Junior developer / DBA |
| Chef de partie / sous chef | Ha | Senior engineer |
| Head chef / Michelin-starred chef | Ri | Architect |
| Fast-food line worker | Skill engineered out of the role | AI-assisted, offshored routine work |
Fast food solved a different problem, and it solved it by deliberately removing that first stage. The sociologist George Ritzer called the pattern McDonaldization: a system optimized for efficiency, calculability, predictability, and control, engineered so a new hire with a few hours of training produces an identical product to someone who has worked the line for five years. That is not a knock on the business model; it works exactly as designed. But it is worth being precise about what it does not produce. Nobody becomes a chef by working a fast-food fryer, however many years they put in, because the job was built not to require or transmit that judgment in the first place. The skill was engineered out on purpose, not lost by accident.
Ready meals sit in between. Reheating something factory-made gets dinner on the table tonight, competently and fast. It teaches nothing about knife work, reduction, or timing, because there was never a decision to make, only a setting to pick. A kitchen that runs entirely on ready meals and fast-food-style assembly can feed people indefinitely. It will never produce a head chef, because nobody in it was ever asked to actually cook.
Why "Just Keep the Masters" Doesn't Work
The appeal of "only the senior experts need to stay in Germany, the routine work can be outsourced or automated" is that it looks efficient in this quarter's budget. It is efficient, right up until the current cohort of seniors retires, changes career, or simply burns out from being the only people left who can do the work nobody trained a replacement for.
Expertise is not a static resource that can be preserved by protecting a headcount line. It is a flow, produced continuously by people doing real work, making real mistakes, and having those mistakes caught and explained by someone more experienced. A senior DBA's judgment about when a maintenance window is too risky, or when a query plan "looks wrong" before the numbers confirm it, was built out of hundreds of smaller, lower-stakes situations early in their career, many of which they got wrong first. Skip that phase and you don't get a faster senior. You get someone who can produce plausible-looking output without the judgment to know when it's wrong, which in database and infrastructure work is precisely the situation that turns a manageable incident into an outage.
The Quiet Failure Mode
This doesn't announce itself as a crisis. It shows up as a slow drift: fewer people in an organization who can explain why a system is built the way it is, more reliance on documentation nobody wrote from lived experience, and a growing gap between "someone who can operate the tool" and "someone who understands the system underneath it." Incident response gets slower, not because the AI-assisted output was wrong, but because there's no longer anyone in the room who can tell, quickly, that it was wrong.
The organizations that will feel this first are the ones that stopped hiring juniors five, ten years ago and assumed the existing seniors were a permanent asset rather than a depreciating one. The next wave of AI-driven "we don't need juniors" decisions is set to make the same bet, on a faster clock, because AI adoption is moving in quarters where the offshoring shift moved in years.
What Would Actually Have to Change
None of this is an argument against AI tools or against global teams. Both are useful and both are staying. The argument is against treating the apprentice stage as pure cost to be minimized rather than as the only mechanism that has ever reliably produced the next generation of Ha and Ri practitioners.
- Protect deliberately low-stakes work for juniors, even when an AI assistant could technically do it faster. The point of the task was never only the output; it was the practitioner learning from doing it.
- Pair juniors with AI tools instead of replacing juniors with them. Reviewing and correcting AI-generated output is itself a Shu-stage exercise, but only if a human is still required to catch what the model gets wrong, not just accept it.
- Treat senior headcount as a pipeline problem, not a retention problem. The question isn't "how do we keep the seniors we have," it's "who becomes the senior after them."
- Stop conflating "cheaper right now" with "sustainable." Both offshoring and AI-driven junior displacement optimize the current budget cycle against a cost that lands five to ten years out, on someone else's watch.
Shu-Ha-Ri survived as a model for centuries because it describes something true about how skill actually forms, not because it's a nice metaphor. Cutting the first stage doesn't produce more masters, faster. It produces a shrinking pool of masters with nobody behind them, and an organization that will eventually notice, at the worst possible moment, that expertise was never something it could just keep on staff. It had to keep being made.