Addy Osmani’s Post

Addy Osmani · article · 2026-07-11 source ↗ Reader 🖥️

AI Employee vs Tool Framing Code Was Never the Bottleneck Agency Human as Orchestrator of Agents Human-in-the-Loop Meta-Skills & Contextual Judgment Taste & Judgment Technical Leadership Track (IC / Staff-Plus Ladder)

As coding agents make solving cheap and abundant, an engineer's durable career value shifts from being the fastest problem-solver to the 'ungradeable' meta-skills that have no answer key — selecting which problems matter, judging quality through earned taste, directing agents rather than doing the work, sprinting the hard last mile, and owning the outcome.

practical · LinkedIn thought-leadership post — a five-point career-advice listicle — A ~288-word LinkedIn post by Addy Osmani titled 'Career advice for engineers in the age of coding agents' (the original post's own slug is 'the-ungradeable-career'). Prescriptive guidance addressed to 'any ambitious engineer navigating the shift', structured as five imperatives plus an opening reframe and a closing synthesis. Practical (tells engineers what to do), not theoretical: it asserts its claims rather than defending them with evidence.

Structure

Part	Role
Opening reframe: the scarce resource has changed	Contrasts the last decade (which rewarded the fastest problem-solver) with this one; argues that when an agent can solve anything 'with an answer key', what is scarce is the judgment to tell which problems agents should solve and whether they got it right.
Move 1 — Define the problem, don't just solve it	Since solving is cheap, selecting becomes scarce: pick which problems are worth the time and frame them precisely enough that an agent can act — 'clear thinking, made legible, is the job.'
Move 2 — Cultivate your taste	Names taste — earned pattern-matching, the ability to tell good from great — as the durable edge; the real risk of AI is losing the ability to tell, so keep your 'alpha ahead of the models' through reps.
Move 3 — Shift from doing to directing	Delegate to an agent as you would to a person: scope the task, define 'done', calibrate trust, verify the result; 'autonomy is a per-task switch, not a rank.'
Move 4 — Sprint the last mile	Agents hand you the first 70-80%; the last 20-30% (edge cases, architecture, polish) is the whole game — 'when first drafts are free, the finish is the product.'
Move 5 — Own what you ship	'The AI did it' is not a defense when prod breaks; be the accountable human who knows exactly what went out the door.
Closing synthesis: focus on the ungradeable part	Ties the five moves together as 'the ungradeable part of the career: choosing what matters, judging honestly when you've got it, and answering for it.'

Key terms

taste Taste & Judgment
Earned pattern-matching — the ability to tell good from great, and more basically to tell whether output is good at all — built through deliberate reps ('study good vs. great'), not innate; the faculty whose loss, not bad code itself, is the real risk of AI.
the ungradeable part
The dimensions of engineering work that have no answer key an agent could check against — choosing what matters, judging honestly whether you've got it, and answering for it; the part of a career that resists automation precisely because it cannot be graded. (Also the original post's slug.)
keep your alpha ahead of the models
Maintain a margin of judgment/skill that exceeds what the models can do, treating your edge as a depreciating asset that must be actively renewed as models improve.
autonomy is a per-task switch, not a rank
How much an agent is allowed to run unsupervised is calibrated task by task (scope, definition of done, trust, verification), not a fixed status granted once and for all.
the last mile
The final 20-30% agents don't hand you — gnarly edge cases, architecture, polish — where difficulty and value now concentrate once first drafts are free.

Propositions

explicitThe scarce resource in the agent era is judgment, not solving: when an agent can solve almost anything given an answer key, what becomes valuable is the judgment to decide which problems agents should solve and whether they actually got it right.conf: high
explicitdevelopedBecause solving is now cheap, the scarce complementary skill is selecting and framing — choosing which problems are worth solving and specifying them precisely enough that an agent can act; 'clear thinking, made legible, is the job.'conf: high
extends ←Stop Just Writing Code. Start Leading Agents: The Staff Engineer’s AI Edge · p8 — Mokhtar's 'move closer to the business' is a concrete instance of the scarce complementary skill of selecting and framing problems.
explicitcorroboratedAn engineer's most durable edge is taste — earned pattern-matching that lets you tell good from great; the real risk of AI is not that it writes bad code but that engineers lose the ability to tell, so judgment must be continually renewed ('keep your alpha ahead of the models') through deliberate reps.conf: high
supportsAgents don't know what good looks like. And that's exactly the problem. · p1 — Prescribes cultivating exactly the professional judgment agents lack: because agents operate at a novice level and cannot tell valid from invalid, human taste is the scarce complement — 'the risk was never AI writing bad code, it's losing the ability to tell.'
supports ←Adam Mosseri: AI is a tailwind for authenticity · p2 — Both name taste as the durable human edge that resists automation.
explicitestablishedEngineers should shift from doing the work to directing agents — delegating as to a person by scoping the task, defining 'done', calibrating trust, and verifying the result — treating autonomy as a per-task switch rather than a fixed rank.conf: high
supportsAI isn't making developers more productive – it's making them busier · p5 — The individual-craft prescription (direct, verify, own) for the measured writing→evaluating role shift.
supports ←11 Tips For AI Coding With Ralph Wiggum · p1 — Osmani's 'shift from doing to directing — scope the task, define done, calibrate trust, verify' is the principle; Ralph is the machinery. The loop is a worked instantiation: the human defines the end state (scope + stop condition), calibrates trust via the HITL->AFK progression, and verifies through commit-blocking feedback loops.
supports ←Stop Just Writing Code. Start Leading Agents: The Staff Engineer’s AI Edge · p2 — Independent convergence on 'direct agents the way you'd delegate to a person' as the core behavioral shift.
supports ←New Skills! v1.1 brings /wayfinder, /research, /implement, /to-spec, /to-tickets · p1 — Osmani's 'shift from doing to directing — scope the task, define done, calibrate trust, verify' is the principle; this lifecycle is the concrete machinery. Grilling/Wayfinder + /to-spec scope the task, tickets and the spec define 'done', the grilling confirmation gate and AFK-vs-human-decision split calibrate trust, and /code-review verifies — a worked instantiation of directing agents rather than doing the work.
explicitValue now concentrates in the last mile: agents deliver the first 70-80%, but the remaining 20-30% — edge cases, architecture, polish — is the whole game, so when first drafts are free the finish becomes the product.conf: high
refinesAccelerating Delivery: How AI Agents Can Own the Initial Code Draft · p1 — Relocates where the value sits: that piece treats automating the ticket→first-PR gap as the productivity win, but Osmani reframes the first draft as the cheap 70-80% — the last mile (edge cases, architecture, polish) is where difficulty and value now concentrate ('when first drafts are free, the finish is the product').
explicitcorroboratedEngineers must own what they ship: '"The AI did it" is not a defense when prod breaks', so remaining the accountable human who knows exactly what went out the door is a core, non-delegable part of the job.conf: high
supportsResearch: Why You Shouldn't Treat AI Agents Like Employees · p6 — '"The AI did it" is not a defense ... your name is on the change' is the individual-ethic form of the finding that today's AI cannot be held accountable, so clear human ownership is always required.
extends ←You Outsourced the AI—but You Still Own the Risk · p1 — Scales the individual 'own what you ship — the AI did it is not a defense' ethic to its organizational/legal form: accountability for outsourced AI is non-delegable and lands on the deploying enterprise, not the upstream builder.
supports ←Should You Still Become a Software Engineer in 2026? GitHub VP · p3 — Both reject 'the AI did it' as a defense when shipped code causes harm.
implicitWhat makes these meta-skills durable is that they are ungradeable: because there is no answer key an agent could check them against, choosing what matters, judging honestly, and answering for the result resist the automation that is commoditizing solvable, checkable work.conf: medium

Arguments

Ambitious engineers should invest in the ungradeable meta-skills — selecting problems, taste, directing agents, finishing the last mile, and owning outcomes — because those are what stay scarce and valuable once agents make solving cheap.

Explicit premises:

When an agent can solve almost anything given an answer key, solving becomes cheap and abundant.
What stays scarce is the judgment to decide which problems to solve and whether the agent actually got it right.
Agents hand you only the first 70-80%; framing the problem and finishing the last mile remain on the human.
AI cannot be held accountable when prod breaks, so a human must own the outcome.

Implicit premises:

Judgment and taste cannot themselves be automated within the relevant time horizon — the durability the piece assumes but never defends.
The value of solving falls faster than the value of the meta-skills rises, so reallocating effort toward the meta-skills is the winning move.
These meta-skills are genuinely learnable and renewable by an individual engineer rather than fixed talent.

Inferences:

If solving is commoditized while framing, judgment, finishing, and ownership are not, an engineer's marginal value migrates to the latter.
Therefore effort should shift from being the fastest solver to being the best selector, judge, finisher, and owner.

Cultivating taste — earned judgment about quality — is an engineer's most durable investment in the agent era.