Agentic coding still needs product taste

Agentic coding changes the speed of implementation, but it does not remove taste.

The fastest draft can still choose the wrong hierarchy, flatten a real workflow into a generic pattern, invent component APIs the team will regret, or pass tests while the product feels off. The tool can write code. It cannot own the promise the product makes to the person using it.

That is why I do not think the strongest AI-assisted engineers are the ones who generate the most code. The strongest ones build a loop around the code: context, intent, constraints, draft, critique, verification, and release proof. They know when to accept generated momentum and when to slow down because the product decision is still unresolved.

For my site, this matters because I want the work to read like someone who can operate in modern engineering teams without outsourcing judgment to the machine.

ContextWhat exists

Repo patterns, product promise, user state, design system, data contracts, and risk.

DraftWhat changed

Generated code, copy, components, migration, tests, and local assumptions.

TasteWhat holds

Hierarchy, scope, system fit, recovery path, and verification that proves the work.

Start with product intent

Before asking an agent to code, I want the product intent written in plain language. The prompt should not only describe files; it should describe the promise being improved.

The questions I would use are:

• What should users understand or do?
• What product pressure created the task?
• Which behavior must remain stable?
• Which tradeoff is acceptable?

The mistake is starting with implementation details before the product decision is clear. That mistake makes the work look finished while hiding the decision that actually matters. It can make a portfolio page louder, a PR harder to review, or a product surface more fragile than it needs to be.

The artifact I want is a prompt header with product intent, invariant behavior, and explicit non-goals. It should be plain enough to inspect and specific enough to be useful. If the artifact cannot show the constraint, the decision, and the proof, the story is probably still too vague.

For AI-assisted implementation where code speed, product judgment, interface taste, and verification all have to stay connected, I want the artifact to be useful before it becomes presentable. It should help someone make a decision, review the risk, or explain the tradeoff without needing a private meeting.

The proof is generated code that can be judged against a real product target. I would rather show a narrow proof that survives questions than a broad claim that only sounds impressive. A hiring manager should be able to ask how I know, what I owned, what changed, and what I would do differently next time.

Feed the local system to the agent

AI output gets weaker when it is asked to invent in a vacuum. The task should include the local component patterns, styling rules, data shape, and testing habits.

The questions I would use are:

• Which components already exist?
• Which helpers are preferred?
• Which design tokens matter?
• Which tests or scripts prove the work?

The mistake is letting the agent create a parallel style because it did not see the system. That mistake makes the work look finished while hiding the decision that actually matters. It can make a portfolio page louder, a PR harder to review, or a product surface more fragile than it needs to be.

The artifact I want is a context bundle that names local patterns before code generation begins. It should be plain enough to inspect and specific enough to be useful. If the artifact cannot show the constraint, the decision, and the proof, the story is probably still too vague.

This is where AI-assisted engineering judgment matters. The work should not depend on taste alone; it should leave a small operating model that another designer, engineer, or reviewer can reuse.

The proof is work that feels native to the codebase. I would rather show a narrow proof that survives questions than a broad claim that only sounds impressive. A hiring manager should be able to ask how I know, what I owned, what changed, and what I would do differently next time.

StateDoes it recover?

Empty, loading, stale, failed, restricted, and partial states.

SystemDoes it belong?

Local components, tokens, helpers, event names, and release habits.

ReaderCan it be reviewed?

Small scope, clear intent, receipts, and a PR story humans can trust.

Keep taste decisions human

The agent can suggest hierarchy and copy, but I want the human to own the final product taste decisions.

The questions I would use are:

• What hierarchy should lead?
• Which state needs the most care?
• What copy sounds like the product?
• What should feel restrained?

The mistake is accepting plausible generated taste because it looks polished enough. That mistake makes the work look finished while hiding the decision that actually matters. It can make a portfolio page louder, a PR harder to review, or a product surface more fragile than it needs to be.

The artifact I want is a human decision note for hierarchy, copy, density, and interaction tone. It should be plain enough to inspect and specific enough to be useful. If the artifact cannot show the constraint, the decision, and the proof, the story is probably still too vague.

For AI-assisted implementation where code speed, product judgment, interface taste, and verification all have to stay connected, I want the artifact to be useful before it becomes presentable. It should help someone make a decision, review the risk, or explain the tradeoff without needing a private meeting.

The proof is a product surface that feels authored instead of assembled. I would rather show a narrow proof that survives questions than a broad claim that only sounds impressive. A hiring manager should be able to ask how I know, what I owned, what changed, and what I would do differently next time.

Use generated critique, not generated confidence

After the first draft, I prefer asking another pass to attack the work. The useful output is not praise; it is missing states and weak assumptions.

The questions I would use are:

• Which state is missing?
• What is vague?
• Where did it ignore local patterns?
• What could regress?

The mistake is asking whether the output is good and receiving generic reassurance. That mistake makes the work look finished while hiding the decision that actually matters. It can make a portfolio page louder, a PR harder to review, or a product surface more fragile than it needs to be.

The artifact I want is a critique prompt that asks for findings with location, impact, and fix. It should be plain enough to inspect and specific enough to be useful. If the artifact cannot show the constraint, the decision, and the proof, the story is probably still too vague.

This is where AI-assisted engineering judgment matters. The work should not depend on taste alone; it should leave a small operating model that another designer, engineer, or reviewer can reuse.

The proof is a stronger second draft with visible review pressure. I would rather show a narrow proof that survives questions than a broad claim that only sounds impressive. A hiring manager should be able to ask how I know, what I owned, what changed, and what I would do differently next time.

Must keepInvariants

Routes, contracts, public copy, accessibility, analytics, and product behavior.

Can changeImplementation

Helpers, structure, naming, tests, and internal composition.

Must proveReceipts

Build, route checks, screenshots, migration checks, and reviewer focus.

Narrow the diff before polish

Agentic coding makes it easy to accumulate adjacent improvements. The diff should be narrowed before anyone debates polish.

The questions I would use are:

• Which files are necessary?
• Which edits are unrelated?
• What can wait?
• What makes review cheaper?

The mistake is shipping a PR that mixes feature, refactor, copy rewrite, and format churn. That mistake makes the work look finished while hiding the decision that actually matters. It can make a portfolio page louder, a PR harder to review, or a product surface more fragile than it needs to be.

The artifact I want is a scope trimming pass before final QA. It should be plain enough to inspect and specific enough to be useful. If the artifact cannot show the constraint, the decision, and the proof, the story is probably still too vague.

For AI-assisted implementation where code speed, product judgment, interface taste, and verification all have to stay connected, I want the artifact to be useful before it becomes presentable. It should help someone make a decision, review the risk, or explain the tradeoff without needing a private meeting.

The proof is a reviewable change that earns trust. I would rather show a narrow proof that survives questions than a broad claim that only sounds impressive. A hiring manager should be able to ask how I know, what I owned, what changed, and what I would do differently next time.

Check production-shaped states

Generated UI often handles the ideal case first. Product taste shows up when the ugly states get the same attention.

The questions I would use are:

• What happens with long content?
• What happens when data is missing?
• Can the user recover?
• Does mobile still work?

The mistake is checking only the screenshot the agent made look clean. That mistake makes the work look finished while hiding the decision that actually matters. It can make a portfolio page louder, a PR harder to review, or a product surface more fragile than it needs to be.

The artifact I want is a fixture set that covers ideal, damaged, empty, restricted, and failed states. It should be plain enough to inspect and specific enough to be useful. If the artifact cannot show the constraint, the decision, and the proof, the story is probably still too vague.

This is where AI-assisted engineering judgment matters. The work should not depend on taste alone; it should leave a small operating model that another designer, engineer, or reviewer can reuse.

The proof is an interface that survives real product pressure. I would rather show a narrow proof that survives questions than a broad claim that only sounds impressive. A hiring manager should be able to ask how I know, what I owned, what changed, and what I would do differently next time.

Make verification part of the prompt

The prompt should name the expected proof before the implementation starts. That keeps the work tied to behavior.

The questions I would use are:

• Which command must pass?
• Which route must be opened?
• Which asset should exist?
• Which migration must be listed?

The mistake is deciding after the fact what proof might be enough. That mistake makes the work look finished while hiding the decision that actually matters. It can make a portfolio page louder, a PR harder to review, or a product surface more fragile than it needs to be.

The artifact I want is a verification checklist inside the task prompt. It should be plain enough to inspect and specific enough to be useful. If the artifact cannot show the constraint, the decision, and the proof, the story is probably still too vague.

For AI-assisted implementation where code speed, product judgment, interface taste, and verification all have to stay connected, I want the artifact to be useful before it becomes presentable. It should help someone make a decision, review the risk, or explain the tradeoff without needing a private meeting.

The proof is a workflow where generated code arrives with an expected QA path. I would rather show a narrow proof that survives questions than a broad claim that only sounds impressive. A hiring manager should be able to ask how I know, what I owned, what changed, and what I would do differently next time.

Write PRs for human reviewers

A PR created with AI should be easier to review, not harder. The description should explain what changed, why it changed, and where the reviewer should look.

The questions I would use are:

• What is mechanical?
• What needs product judgment?
• What was verified?
• What risk remains?

The mistake is assuming the reviewer will infer intent from a generated diff. That mistake makes the work look finished while hiding the decision that actually matters. It can make a portfolio page louder, a PR harder to review, or a product surface more fragile than it needs to be.

The artifact I want is a reviewer-focus block in every AI-assisted PR. It should be plain enough to inspect and specific enough to be useful. If the artifact cannot show the constraint, the decision, and the proof, the story is probably still too vague.

This is where AI-assisted engineering judgment matters. The work should not depend on taste alone; it should leave a small operating model that another designer, engineer, or reviewer can reuse.

The proof is a PR that respects human review time. I would rather show a narrow proof that survives questions than a broad claim that only sounds impressive. A hiring manager should be able to ask how I know, what I owned, what changed, and what I would do differently next time.

Keep the paper trail small

Receipts do not need to become bureaucracy. They need to be small enough to maintain and specific enough to matter.

The questions I would use are:

• What did the agent do?
• What did the human decide?
• What proof exists?
• What follow-up remains?

The mistake is creating giant process notes no one will read. That mistake makes the work look finished while hiding the decision that actually matters. It can make a portfolio page louder, a PR harder to review, or a product surface more fragile than it needs to be.

The artifact I want is a four-part receipt: agent task, human decision, verification, follow-up. It should be plain enough to inspect and specific enough to be useful. If the artifact cannot show the constraint, the decision, and the proof, the story is probably still too vague.

For AI-assisted implementation where code speed, product judgment, interface taste, and verification all have to stay connected, I want the artifact to be useful before it becomes presentable. It should help someone make a decision, review the risk, or explain the tradeoff without needing a private meeting.

The proof is evidence that survives later questions. I would rather show a narrow proof that survives questions than a broad claim that only sounds impressive. A hiring manager should be able to ask how I know, what I owned, what changed, and what I would do differently next time.

Show the loop in candidate work

The strongest portfolio angle is not that I used AI. It is that I built a reliable human-in-the-loop workflow around it.

The questions I would use are:

• What did AI accelerate?
• Where did I correct it?
• What artifact proves judgment?
• What shipped safely?

The mistake is presenting AI work as magic output without showing taste or ownership. That mistake makes the work look finished while hiding the decision that actually matters. It can make a portfolio page louder, a PR harder to review, or a product surface more fragile than it needs to be.

The artifact I want is a case-study sidebar showing prompt, critique, accepted change, rejected change, and QA receipt. It should be plain enough to inspect and specific enough to be useful. If the artifact cannot show the constraint, the decision, and the proof, the story is probably still too vague.

This is where AI-assisted engineering judgment matters. The work should not depend on taste alone; it should leave a small operating model that another designer, engineer, or reviewer can reuse.

The proof is a candidate signal that matches modern engineering reality. I would rather show a narrow proof that survives questions than a broad claim that only sounds impressive. A hiring manager should be able to ask how I know, what I owned, what changed, and what I would do differently next time.

What I would show in the work

The public version should show the working artifacts, not only the final opinion. For AI-assisted implementation where code speed, product judgment, interface taste, and verification all have to stay connected, I would include the matrix, the state map, the review checklist, and the before-and-after decision path. Those artifacts make the work feel authored because they reveal how the decision was made.

I would also include what I did not do. That is often where judgment is clearest. Not every useful idea belongs in the first version. Not every dashboard needs live sync. Not every component needs a new prop. Not every AI suggestion belongs in the PR. Naming the boundary helps the reader trust the result.

The page should make the work inspectable without turning into internal documentation. I want enough specificity for an engineering manager to ask serious follow-up questions, and enough restraint that the story still reads like product judgment instead of a dump of process artifacts. The best version makes the artifacts feel inevitable: this was the pressure, this was the decision, this was the receipt, and this is why the outcome is believable.

RejectToo broad

The agent expanded scope, invented patterns, or ignored local constraints.

RefineAlmost right

The draft works but needs hierarchy, copy, states, or system alignment.

ShipProved

The change is scoped, reviewed, verified, and connected to product intent.

Downloadable companion

This topic deserves a companion resource: an agentic coding taste checklist with intent, constraints, states, local patterns, verification, and reviewer notes. It should be useful as a working file, not a decorative download. The resource should help someone repeat the review, pressure-test the decision, and carry the same quality bar into their own product work.

I would keep it concise: one page if possible, with fields for context, constraint, decision, evidence, owner, and follow-up. The value is not the file format. The value is that the artifact turns the article into something someone can use.

Review checklist

Before publishing this work, I would run a short review against the same standard I use for product changes:

• Is the product pressure concrete?
• Is my ownership clear?
• Is the system constraint named?
• Is there at least one artifact that proves the decision?
• Does the artifact show a real tradeoff?
• Is the metric or signal honest about its limits?
• Are support, operations, accessibility, or release risks named when relevant?
• Does the writing explain what I intentionally left out?
• Can a recruiter skim the point quickly?
• Can an engineer ask a deeper technical question?
• Does the downloadable resource make the idea reusable?
• Would I be comfortable defending the claim live?

That checklist keeps the work from becoming a polished but vague page. It also protects the voice. The goal is not to sound like a process manual. The goal is to make the product judgment visible enough that a hiring team can trust the story.

Implementation notes

The implementation version of this idea should be small enough to ship and specific enough to prove. I would start by naming the route, artifact, owner, and verification path before adding polish. If the work touches content, I would check the source body, generated route, metadata, sitemap, and social image. If it touches UI, I would check desktop, mobile, long content, empty state, keyboard path, and the most likely failure state. If it touches data, I would name the source of truth, freshness, migration path, and what support or product should see after launch.

That implementation note matters because AI-assisted engineering judgment can drift when the work moves from idea to code. A good article can describe the principle, but a good product change needs the boring details: filenames, states, commands, rollback, ownership, and the reason the first version is intentionally narrow.

I would also write the follow-up before shipping. Follow-up is not a sign that the work is incomplete; it is a sign that the boundary is known. The first version should solve the risky problem, prove the pattern, and leave the next step visible. That is how small teams move quickly without pretending every release is final.

For portfolio proof, these implementation notes are useful because they make the story harder to fake. They show that I understand the difference between a good idea, a shippable version, and a maintainable system. They also give an interviewer concrete places to dig: why this scope, why this artifact, why this verification path, and what changed after the first release.

Case-study packaging

If this became a Work section detail, I would package it as a small evidence stack. The top should explain the product pressure in plain language. The middle should show the artifact and the operating decision it supported. The bottom should show the verification and the follow-up. That structure keeps the story from becoming either a pretty screenshot or a private engineering note.

The captions matter here. A caption should not say "dashboard view" or "component states" and stop there. It should explain what the reader is supposed to learn: this matrix shows why the first version stayed narrow, this state map shows where recovery mattered, this QA note shows how the release was proved, or this event taxonomy shows how product language became measurable.

I would keep the packaging honest by including one caveat. The caveat might be a metric limitation, a data freshness issue, a rollout boundary, a support dependency, or a follow-up that intentionally stayed out of scope. That caveat does not weaken the case study. It makes the judgment feel real.

The final test is whether the page creates a better conversation. If the artifact helps someone ask a sharper question about product judgment, implementation detail, or release proof in real live interviews together, it belongs in the story.

Interview angle

In an interview, I would explain this through agentic coding as a product craft loop, not an outsourcing trick. The story should start with the product pressure, then move into the system constraint, the artifact, and the proof. That order keeps the answer grounded. It also gives the interviewer several places to go deeper: data, frontend architecture, design systems, support, migration, accessibility, or release process.

The strongest version of the answer includes a tradeoff. I want to be able to say what I chose, what I left alone, and how I knew the work helped. That is more credible than presenting every project as a clean win.

The hiring signal

Agentic coding with product taste is a hiring signal because it shows I can use AI to move faster without losing the human judgment that makes product work worth shipping.

That is the level I want this site to communicate. The work should show taste, but it should also show operating judgment. It should make me look like someone who can enter a real product system, understand the messy middle, ship the useful version, and leave enough proof for the next person to trust it.