Boundary-Crossing 2026

The wall became a choice.

Engineers doing design, server engineers doing client, planners writing code. Crossing job-function boundaries is rising, because AI made the execution of each domain cheap.
But the gate to crossing isn't execution. It's the taste to judge what's good in the next field, and taste is something AI doesn't hand you. Neither broad wins nor deep wins on its own — here is the craft of crossing in the AI era.

59/100
workers will need re/upskilling by 2030 (WEF 2025; 19 of them redeployed to a different role within their org)
39%
of core skills change by 2030 — and falling (57 in 2020 → 44 in 2023 → 39) (WEF 2025)
2x
citation effect of "a deep conventional base + a measured atypical injection" (Uzzi, Science 2013)
Apr 2025
Shopify makes AI usage a baseline + a hiring gate (Lütke)
↓ SCROLL
01 / The Thesis

When execution gets cheap,
judgment becomes the scarce thing.

Why is crossing possible now? Because AI made the execution of each domain cheap. Execution can be delegated. What you can't hand off is the breadth to judge whether the delegated work is any good. This isn't new — entrepreneurship research formalized it over twenty years ago.

Lazear / 2002

The crosser must "judge the quality"

Verbatim: "an entrepreneur must be sufficiently well-versed in a variety of fields to judge the quality of applicants." Execution can be hired (today: delegated to AI). What can't be given up is the breadth to judge. The classic scaffolding for "the gate to crossing is taste."
Balanced Skills and Entrepreneurship (NBER 9109)
Field reality / 2026

You can "make it." Whether it's good — unclear

You can have AI produce the design, the code, the spec. But if you can't judge whether the output is good, crossing stalls at "looks-done." AI fills generation. It doesn't fill judgment.
The central question of this report
Series link

Execution cheap, judgment scarce

Same through-line as the prior report on decision-making. The talent-side sequel to Prediction Machines (prediction/building gets cheap, judgment becomes the scarce complement). Crossing means holding that scarce judgment across multiple domains.
Connects to ai-era-decision-making

How to read this report (what's proven, what's a frame)

Let's be honest about the line. "Judgment is the gate" is supported by Lazear's classic theory and the academic work below. But the causal claim that "AI made execution cheap, therefore crossing is newly feasible" is not established by primary sources in this research (the view of Netflix's role-collapse as an AI precursor was rejected in verification). Read it as the series' framing plus an observable phenomenon (Shopify, vibe coding). The "crossing distance" model below is also an organizing frame, not a proven law.

→ The numbers and quotes are checked facts. The causal arrows and the model are plausible frames, read as such.

02 / The Lens

Crossing has
a distance.

Lump "crossing" together and you miss the point. Stepping into the next function differs from moving to another industry. Split it into three by distance. The farther you go, execution gets filled by AI but the taste gap widens — this is not evidence, it's this report's organizing frame.

Near

Across functions

Engineer ↔ design ↔ PM ↔ QA. The next role inside the same product. Lots of shared context, and taste is easy to borrow. The most feasible, and where it's actually happening.

Mid

Across domains

front ↔ back ↔ infra ↔ data ↔ ML. The neighboring discipline. AI fills the syntax, but what counts as "good design" differs by domain. You have to import the taste.

Far

Industry / career

A different industry or a career switch. AI accelerates learning and execution, but the sense of what's "good" in that industry is the hardest to acquire. Maximum distance, maximum taste gap.

03 / Current State

The world already
assumes crossing.

Crossing is moving from "nice to do" to "assumed." Both in top-down mandates and in the macro numbers.

Shopify / Apr 2025

AI usage is a baseline, hiring has a gate

CEO Tobi Lütke told the whole company (2025-04-07): "Reflexive AI usage is now a baseline expectation at Shopify." Plus a hiring gate: "Before asking for more headcount, demonstrate why you cannot get it done using AI." AI fluency factors into reviews. A concrete instance of mandating that everyone cross into AI-augmented adjacent work.
Corroborated by CNBC; Forrester flags enforceability
WEF Future of Jobs / Jan 2025

59 of every 100 will re-learn

By 2030, 59/100 need re/upskilling. The split: 29 upskilled in role, 19 redeployed to a different role within the org, 11 unlikely to get the training (= 120M+ at medium-term redundancy risk). That "19 redeployed" is intra-org boundary-crossing. 39% of core skills change by 2030 (though falling: 57 → 44 → 39).
Self-reported projections from 1,000+ employers
What's demanded is the "combination"

Per WEF 2025, the fastest-growing skill is AI and big data; the #1 core skill is analytical thinking (70% of employers call it essential). And it states that "a combination of both [technical and human] skillsets will increasingly be required" across growing jobs. Yet the same report also stresses "domain expertise that cannot easily be automated" (i.e., depth). Testimony that the answer is neither breadth alone nor depth alone.
Note: WEF measures skill instability / re-skilling, not boundary-crossing itself; the bridge to crossing is this report's interpretation. Figures are as of January 2025, horizon 2030.

04 / Prior Art

The words for crossing
already exist.

Crossing is a new phenomenon, but the tools to think about it were built by others. Here is the lineage, dated precisely (correcting a common error).

Concept
Proposer / Year
In one line
T-shaped peoplevertical = depth + horizontal = crossing
Johnston 1978 / Guest 1991
Note: HBR 2001 is the "T-shaped MANAGER (mgmt application)," not the term's origin
Jack-of-all-trades (balanced skills)breadth predicts entrepreneurship
Edward Lazear / 2002
★The gate is the breadth to judge quality; execution can be delegated
Glue workthe non-code crossing/coordination labor
Tanya Reilly / 2018-2019
Makes teams succeed, yet undervalued at promotion
Full cycle developerdesign through operate, one team
Netflix / 2018
Role-collapse of SWE/SDET/SRE (NOT an AI precursor)
Range (the case for breadth)defense of generalists
David Epstein / 2019
Related book (outside this report's verification batch; reference)
05 / The Economics

"Broad wins" and
"deep wins" are both wrong.

The economics of crossing is well studied. The verdict is neither blanket praise nor contrarian rejection. The shape that works is asymmetric: a deep base, plus one measured crossing on top.

EVIDENCE 01 — Uzzi et al. (Science 2013)
novelty aloneconvention + one atypical hit

An analysis of 17.9 million papers. The highest-impact work is grounded in conventional combinations, with a single intrusion of an unusual combination. That profile was twice as likely to be highly cited. Breadth/novelty "alone" doesn't do it.

Note: measured via journal-pair co-citation within papers, not an individual's T-shape depth. This report uses it by analogy (correlation).

EVIDENCE 02 — Mannucci & Yong (AMJ 2018)
depth is always gooddepth is stage-dependent

A longitudinal study of 2,070 animators. Depth helps creativity early in a career, but turns into a liability late (creativity −7% to −24%); breadth helps late. So "keep thickening the vertical bar" is not optimal. When you start crossing matters.

The claim that depth must mechanistically precede breadth was rejected in verification. This is an observed reversal of effect by stage, not a tested causal sequence.

Lazear panel

Breadth predicts "crossing"

In a panel of ~5,000 Stanford GSB alumni, the number of prior roles is the single best variable explaining who starts a business (t-value near 30). But it's cross-sectional, so it predicts, it doesn't "drive" (cause).
Astebro & Thompson (2011)

But "broad = success" isn't a given

In a survey of 830 inventors and others, varied work history correlated with lower household income (possibly a "taste for variety" selection effect). Whether breadth causes success or just reflects a preference is unresolved. Don't over-praise it.
06 / The Trap

Shallow crossing
becomes debt.

Crossing isn't free. Reach wide without judgment and a quiet tab of quality and coordination debt piles up. Three traps.

1. The "looks-done" problem

AI produces a plausible-looking artifact in the next domain. But if you can't judge quality, you won't catch the flaws. Generation got cheaper; verification did not.

→ Until you can tell "good" in the new domain, make a specialist's review a condition of done.

2. The glue-work trap

Crossing and coordination work (Tanya Reilly's glue work) earns praise in reviews yet is undervalued at promotion ("no visible technical contribution"). Crossing too early can stall a career.

→ Counter-view: Sean Goedecke reframes it as a rational company trade-off. Read it as: unless the org rewards it structurally, the individual loses.

3. The drop-your-depth trap

Declare "it's all breadth now" and let go of the vertical bar, and you get crossing with no base. Uzzi shows a conventional base is the premise; Lazear says breadth-to-judge is required. Breadth only works on top of a base.

→ Keep one deep bar. But per Mannucci, shift the ratio with career stage.

4. The stay-narrow-on-execution trap

If your value is "I build it fast and right," that sits on exactly what AI cheapens most. Now that execution is cheap, a small team that crosses several domains — a startup, or a lean elite squad inside a big company — outruns a large org split into narrow specialists. The moment you narrow your role to execution, you become the replaceable part.

→ Narrow your taste, not your execution. Move to the side that judges good from bad in the domain.

07 / Playbook

Crossing with AI:
the craft.

How the individual moves, how the organization enables it. Guidance drawn from the evidence.

Individual: when you cross
Why
Learn "what good looks like" before the syntax
AI fills execution. What you can't hand off is taste (Lazear). Soaking in good and bad examples of the next domain is the real act of crossing.
AI drafts, a specialist reviews
Generation is cheap, verification is expensive. Make review part of the definition of done to stop "looks-done."
Reversible: do it yourself. Irreversible: defer
Cross fast yourself on reversible calls; defer irreversible ones to a specialist (Bezos's reversibility heuristic; primary support is thin, so treat it as guidance).
Keep one deep bar, but shift the ratio by stage
A deep base plus a measured crossing wins (Uzzi). Depth thick early, lean toward breadth later (Mannucci).
Organization: when you enable crossing
Why
Reward crossing and coordination at promotion
If glue work goes unrewarded, the people who can cross leave. Unless it's made visible in the rubric, the individual loses (Reilly).
Keep specialists as "taste reviewers"
Even when everyone crosses, without a gatekeeper of "good" per domain, quality collapses. The specialist's role shifts from execution to judgment.
Design redeployment around crossing
WEF's "19/100 redeployed within the org" is crossing itself. Build the landing pad for re-skilling and you shrink the at-risk 11.
Pair "just use AI" mandates with a taste guarantee
Shopify-style pressure accelerates crossing, but without a judgment mechanism it mass-produces "looks-done". Mandate and quality gate go together.
08 / Key Messages

People remember "lines,"
not "frameworks."

Lines that land — for meetings, slides, or a single sentence.

AI made the execution of other people's jobs cheap. So the wall became a choice — cross it or not.
The gate to crossing isn't execution. It's whether you can tell what's "good" in the next field. AI fills generation. It doesn't fill taste.
"Broad wins" is a lie. A deep base, plus one measured crossing on top — that's the shape that works (Science 2013).
An org that doesn't reward crossing at promotion loses the people who can cross.
When execution is cheap, a small team that crosses outruns a big org split into specialists.

Sources

Primary/secondary sources that passed adversarial verification. Correlation is flagged as correlation, and claims that couldn't be substantiated (Netflix as an AI precursor, a mechanical depth→breadth causation, a decisive breadth→low-income reading) are not used as support. Numbers are identical across JA/EN.

More in this series — AI Era Insights
Decision-Making
Organizational Decision-Making in the AI Era
Hiring Strategy
The Talent Blueprint for the AI Era
Light & Shadow
Light and Shadow of the AI Era