PRODUCT.ENGINEER
ManifestoThe RolePlaybookLoops
Back to blog
engineeringAugust 5, 202619 min read

Software Craft 2026: Why Taste Is the Only Advantage Left

Software craft 2026 is defined by taste, not speed. When AI builds anything, product engineers with judgment make the difference.

Felipe Barreiros

On this page

  • The cursor blinks. So what?
  • Software craft 2026: a definition
  • The taste spectrum: what separates great from adequate
  • Why taste became the differentiator
  • The craft framework: four dimensions of taste
  • Measuring craft: the signals that matter
  • The product engineer as taste-keeper
  • How to cultivate taste in engineering teams
  • The organizational cost of taste
  • Software craft 2026 in practice: a day in the life
  • The market reward for craft
  • The AI collaboration model for craft
  • Key takeaways
  • FAQ
  • Related reading

On this page

  • The cursor blinks. So what?
  • Software craft 2026: a definition
  • The taste spectrum: what separates great from adequate
  • Why taste became the differentiator
  • The craft framework: four dimensions of taste
  • Measuring craft: the signals that matter
  • The product engineer as taste-keeper
  • How to cultivate taste in engineering teams
  • The organizational cost of taste
  • Software craft 2026 in practice: a day in the life
  • The market reward for craft
  • The AI collaboration model for craft
  • Key takeaways
  • FAQ
  • Related reading

The cursor blinks. So what?

Software craft 2026 faces a strange paradox: you can build anything now, yet most of what gets built is mediocre. A full SaaS product in a weekend. A mobile app before lunch. An internal tool between meetings. The cursor blinks and the machine fills the screen with working code, correct types, passing tests. The generation problem is solved. And yet most software feels the same as it did a decade ago. Functional but forgettable. Present but not purposeful.

At product.engineer, we define software craft in 2026 as no longer about capability but about the willingness to care when caring is optional. It is taste applied at every layer of the stack, from the database schema to the loading state animation to the exact moment a toast notification disappears. A product engineer embodies this principle: they own the full outcome, not just the implementation, and they bring judgment to places where the machine brings only output.

Join 2,000+ engineers who define, build, and ship.

One email per week. Practical frameworks for product engineers. No spam.

Tuomas Artman, CTO of Linear, articulated this tension in his AI Engineer talk that accumulated over 9,900 views on YouTube. His thesis was direct: when the cost of building approaches zero, taste becomes the only competitive advantage. Not architecture. Not scaling. Not even speed to market. Taste. The human capacity to look at something technically correct and say "no, this is not good enough for the people who will use it."

The question for every engineering team is not "can we build this?" It is "do we have the judgment to build this well?" This article explores what that judgment looks like in practice, where it comes from, and why the teams that cultivate it are pulling away from everyone else.

Software craft 2026: a definition

Software craft in 2026 is the disciplined application of taste, judgment, and technical excellence to every decision in the product development lifecycle. It is not perfectionism. It is not gold-plating. It is the systematic refusal to ship work that does not meet a standard you can articulate and defend.

This definition matters because the term "craft" has been diluted. For decades it meant writing clean code, following patterns, maintaining test coverage. Those things still matter, but they are now table stakes. AI generates clean code. AI follows patterns. AI writes tests. Craft in 2026 means the things AI cannot do on its own: choosing what to build, deciding how it should feel, knowing when to stop adding features, and understanding why a particular interaction should take 200ms instead of 50ms or 800ms.

Linear is the canonical example. Their product does what every project management tool does: tracks issues, manages sprints, visualizes workflows. Jira does it. Asana does it. Monday does it. But Linear feels different. Every animation is intentional. Every keyboard shortcut is discoverable at the exact moment you need it. The command palette responds in under 16ms. These are not product decisions made in a planning meeting. They are craft decisions made by engineers who cared about the texture of every interaction.

Linear raised their Series C at a $1.25B valuation in 2025 while operating profitably with a deliberately small team and over 15,000 paying companies. For comparison, Atlassian reached a similar valuation only after scaling to thousands of employees. The gap is not explained by features. Jira has more features. The gap is explained by craft. Linear built less, but built it with such intentionality that teams pay premium prices and switch from entrenched competitors with functioning integrations and established workflows.

The taste spectrum: what separates great from adequate

According to product.engineer's framework for craft, taste in software is not subjective. It is a learned skill with identifiable components. Here is how it manifests across the stack:

LayerAdequate (no taste)Good (some taste)Great (deep craft)
API DesignReturns data, correct typesConsistent naming, paginationPredictable patterns, evolves without breaking, self-documenting
Error HandlingShows error messageCategorizes errors, actionable copyAnticipates errors, prevents them, recovers gracefully
Loading StatesSpinnerSkeleton screensProgressive disclosure, instant optimistic updates, no perceived wait
Data ModelStores everything neededNormalized, indexedShaped for the access patterns of the actual product, not hypothetical ones
InteractionsClick triggers actionSmooth transitionsKinetic, responsive, communicates system state through motion
OnboardingDocumentation linkGuided tourZero-instruction comprehension, progressive complexity

The great column is where craft lives. Notice that none of it is about technical difficulty. A skeleton screen is not harder to build than a spinner. Optimistic updates are not orders of magnitude more complex than loading states. The difference is that someone decided to care. Someone looked at the standard implementation and asked "what would make this feel right?" That someone is almost always a product engineer, someone who thinks about the person on the other end of the screen.

Why taste became the differentiator

Three forces converged to make taste the primary competitive axis in software:

1. Generation costs collapsed

GitHub reported in their 2025 Octoverse analysis that 41% of code in new repositories was AI-generated, up from 22% the previous year. Building became trivially easy. A solo developer with good prompting skills can produce the output of a five-person team from three years ago. Output volume stopped being a differentiator. The constraint moved upstream.

2. Users developed higher expectations

Stripe, Linear, Figma, and Vercel spent years training users to expect software that feels considered. When you use Linear's keyboard navigation and then switch to a competitor, the competitor feels broken even if it is technically functional. These products created a new baseline. Users now notice when software lacks polish, even if they cannot articulate why.

3. Distribution fragmented

In a world with three project management tools, you could win on features. In a world with three hundred, you win on experience. The explosion of AI-built tools means users have more choices than ever. They leave software that feels generic because switching costs dropped alongside building costs. If a competitor can rebuild your feature set in weeks, the only durable advantage is taste.

These forces created the current moment. The teams that invest in taste are building moats. The teams that optimize only for output are building commodities.

The craft framework: four dimensions of taste

After years of studying high-craft teams at companies like Linear, Stripe, Vercel, and Figma, and from my own experience building products as a Sr. Product Engineer at AWS and through two startups, I have identified four dimensions that consistently separate great software from merely functional software:

Dimension 1: Intentional reduction

The first sign of craft is what is absent. Great software says no to features that adequate software says yes to. Linear launched without time tracking. Stripe launched without a dashboard. Figma launched without a component library. Each of these decisions was counterintuitive at the time and correct in retrospect. They forced the teams to nail the core experience before adding complexity.

The product engineer's role here is critical. They understand the user well enough to know what can wait. They have the technical judgment to know what architectural decisions a premature feature would force. And they have the conviction to defend a smaller scope against stakeholders who equate features with progress. As I discuss in the product engineering culture article, this kind of ownership requires organizational structures that trust engineers with product decisions.

Dimension 2: Interaction fidelity

Interaction fidelity means that every micro-interaction communicates system state accurately and feels proportional to its importance. A save action that takes 50ms should feel instant. A save action that takes 3 seconds should communicate progress. A destructive action should create momentary friction. These are not product requirements. They are craft standards.

Vercel's deployment interface is a masterclass in interaction fidelity. When you push code, the build logs stream in real-time with syntax highlighting. The deployment URL becomes clickable at the exact moment the deployment is ready. The status transitions feel alive, not like a database being polled. None of this is technically groundbreaking. All of it requires someone who cares about how the deployment experience feels.

Dimension 3: Coherence across layers

Craft manifests when the API, the data model, the business logic, and the interface feel like they were designed by a single mind with a consistent philosophy. Incoherent software reveals its seams: the API uses camelCase but the database uses snake_case and the error messages use a different vocabulary than the UI copy and the loading states vary wildly between pages.

Shopify's Admin API exemplifies coherence. Every resource follows identical patterns. Pagination works the same everywhere. Error responses share a structure. This is not accidental. It requires investment from engineers who understand that consistency is a feature, perhaps the most important one for developers building on your platform.

Dimension 4: Temporal awareness

Great software accounts for time. Not just performance (though that matters), but the understanding that users exist in time and their context changes. A notification that arrives 30 seconds after you already found the information yourself is worse than no notification. A confirmation dialog that appears after you have already moved on mentally is an interruption, not a safeguard. An undo option that expires after 5 seconds is theater.

Notion handles temporal awareness well in their real-time collaboration features. When a collaborator edits a block you are reading, the change appears gradually rather than replacing content abruptly. This is a craft decision that requires someone to think about the user's experience over seconds and minutes, not just the correctness of data synchronization.

Measuring craft: the signals that matter

You cannot improve what you cannot measure. Here are concrete signals that indicate craft quality:

  • Time to value for new users: how many seconds between signup and the first moment of genuine usefulness. Linear achieves this in under 30 seconds for users migrating from other tools.
  • Support ticket vocabulary: when users say "it just works" or describe the product in emotional terms, craft is present. When they file tickets describing confusion, craft is absent.
  • Feature adoption without documentation: the percentage of features users discover and use without reading docs or watching tutorials. High adoption without education signals intuitive design.
  • Retention at 90 days: products with deep craft retain users because switching to alternatives feels like a downgrade, regardless of feature parity.
  • Error rate in production: not just crashes, but soft errors. Confused users, abandoned flows, retried actions. Great craft anticipates these paths.

PostHog publishes their metrics publicly and reported a 90-day retention rate of 62% for activated users in Q4 2025. The median B2B SaaS 90-day retention hovers around 35-40% according to Mixpanel's 2025 Product Benchmarks report. PostHog attributes this gap to their investment in product experience: session replay loads instantly, dashboards configure without documentation, and feature flags deploy in a flow that feels like a single action.

The product engineer as taste-keeper

The product engineer is uniquely positioned to be the guardian of craft because they operate across the boundary that traditionally separates "what to build" from "how to build it." Traditional organizations separate design, product, and engineering into distinct functions with handoff points. Taste dies at handoff points. It cannot survive translation from Figma to Jira to pull request.

When they encounter a loading state, they do not implement whatever the design spec shows. They ask whether a loading state should exist at all. Can the data be prefetched? Can the interaction be optimistic? Can the perceived wait be eliminated by restructuring the flow? These are not engineering questions or design questions or product questions. They are craft questions. And they require someone with the authority and ability to ask them.

In my career, having hired over 600 engineers and coached 12,000 more, the pattern is unmistakable. Engineers who produce great software are not necessarily more technically skilled than their peers. They are more opinionated about quality. They have a visceral reaction to mediocrity that compels them to iterate past "good enough." Cultivating this sensibility is the challenge of engineering leadership in 2026.

This connects directly to what I explore in building in a world of slop: when the floor drops out, when generating acceptable output becomes trivial, the only remaining differentiator is the ceiling. Taste is the ceiling. In product.engineer's Quality Stack, the top two layers, Coherence and Craft, are explicitly marked as "human-required" because they represent exactly this kind of judgment that no automation can replicate.

How to cultivate taste in engineering teams

Taste is learned, not innate. Here are practices that develop it:

1. Use great software constantly and critically

Engineers who use Linear, Figma, Stripe, and Arc Browser daily internalize patterns of quality. Not just by using them, but by studying them. Why does this animation exist? What would this interaction feel like without the 100ms delay? Why does this empty state show this specific illustration? Deliberate observation of craft decisions builds the mental library that taste draws from.

2. Ship to yourself first

The best craft decisions come from teams that use their own product obsessively. Stripe engineers process payments through Stripe. Vercel engineers deploy on Vercel. PostHog engineers analyze their own product with PostHog. When you are the user, you notice the paper cuts that external feedback cycles miss.

3. Slow down at the edges

The core feature deserves speed. The edges deserve slowness. Error states, empty states, loading states, transitions between states. A team that spends 60% of implementation time on the happy path and 40% on the edges will produce software that feels dramatically better than a team that allocates only 10% to everything outside the golden path.

4. Create taste artifacts

Document your opinions. What does good look like in your codebase? How fast should interactions feel? What is your error-handling philosophy? Teams that articulate their taste standards can maintain them as they grow. Teams that keep taste implicit lose it the moment the original craftspeople leave.

5. Review for feel, not just correctness

Code review typically checks: does it work, does it follow patterns, does it have tests. Craft-focused teams add: does it feel right? Would I want to use this? Is there a moment in this flow that will frustrate a real person? These questions are subjective, but that is the point. Objective criteria produce objectively adequate software. Subjective judgment produces great software.

The organizational cost of taste

Craft is not free. Linear ships less frequently than some competitors. Stripe's API takes longer to evolve. Figma spent years on a file format before launching multi-player. The engineer who insists on getting a loading state right is blocking the next feature for an afternoon.

The question is whether the investment compounds. The data says it does. Linear's revenue growth with 80 employees suggests craft is not a handicap. Stripe processes over $1 trillion annually with a product developers choose over alternatives with more features.

The cost of taste is real but bounded. The cost of tastelessness is exponential: churn, support burden, competitive vulnerability, and the erosion of team morale as engineers realize they are shipping work below their capability.

OpenAI learned this publicly. ChatGPT's early success was partly due to interaction design that competitors took months to replicate: the streaming text response, the clean interface, the absence of enterprise chrome. Taste bought them time that engineering alone could not.

Software craft 2026 in practice: a day in the life

What does craft look like on a Tuesday afternoon? An engineer needs to add a confirmation dialog for bulk deletion.

The tasteless approach: modal with "Are you sure?" text and two buttons. Done in twenty minutes.

The craft approach: question whether a modal is the right pattern at all. Modals interrupt flow. Could this be an undo instead? If undo, what is the time window? Five seconds is too short if someone realizes the mistake while getting coffee. Maybe the answer is a two-phase approach: immediate soft delete with a 30-second undo, then permanent deletion after 24 hours with a recovery option in settings.

That is not twenty minutes. That is an afternoon. But the engineer with taste does not just implement the requirement. They interrogate it. They ask what a thoughtful friend would build if they genuinely cared about the person using it. That is the difference between software that works and software that is great.

The market reward for craft

Venture capital follows craft when craft produces retention. Companies with exceptional net revenue retention share a common thread: users describe the product in terms of how it feels, not just what it does. "Fast." "Clean." "Just works." These are the words users employ when someone with taste shaped their experience.

The market reward is also visible in hiring. Linear attracts engineers from companies three times their size because people want to work on products they are proud of. PostHog's community grows because contributors feel the quality bar and rise to meet it.

Software craft in 2026 is a compounding asset. Every craft decision makes the next one easier because the team develops shared vocabulary. Every tasteless shortcut makes the next one more likely because the bar drops imperceptibly with each compromise.

The AI collaboration model for craft

This article is not anti-AI. AI is the most powerful tool for craft we have ever had, when directed by someone with taste. The product engineer in 2026 uses AI to handle the commodity work, the boilerplate, the repetitive patterns, the initial implementations, freeing their attention for the craft decisions that matter.

The model looks like this:

  1. AI generates: the initial implementation, the scaffold, the first pass
  2. Human evaluates: does this meet the craft standard? Does it feel right? Is anything missing from the edge cases?
  3. AI iterates: based on specific, opinionated direction from the human
  4. Human refines: the final 20% that requires taste, the timing, the feel, the coherence

This is not a workflow where AI replaces judgment. It is a workflow where AI amplifies craft by handling the parts that do not require taste. The engineer becomes an editor and director rather than a typist.

The teams that understand this model are producing remarkable software at remarkable speed. The teams that treat AI as a replacement for taste are producing more slop, faster. As the state of AI code quality makes clear, the tooling does not determine the outcome. The person directing the tooling does.

Key takeaways

  • Software craft in 2026 is the human judgment that transforms technically correct software into genuinely great software.
  • Taste is the only remaining competitive advantage when AI commoditizes implementation speed for everyone equally.
  • Teams that treat AI as a replacement for taste produce more slop faster; teams that direct it with taste produce remarkable software.
  • Craft encompasses interaction design, coherence across layers, temporal awareness, and intentional reduction of complexity.
  • The person directing the tooling determines the outcome, not the tooling itself.

FAQ

What is software craft in 2026?

Software craft in 2026 is the disciplined application of taste, judgment, and technical excellence to every decision in the product development lifecycle. It goes beyond clean code and test coverage to encompass interaction design, coherence across layers, temporal awareness, and the intentional reduction of unnecessary complexity. It is the human judgment that transforms technically correct software into genuinely great software.

How is taste different from design?

Taste operates at every layer of the stack, not just the visual interface. A tasteful API is as important as a tasteful onboarding screen. Taste informs database schema decisions, error-handling philosophy, and deployment workflows. Design is one expression of taste. Taste is the underlying judgment that produces good design, good architecture, and good product decisions simultaneously.

Can AI replace taste in software development?

No. AI generates outputs based on patterns in training data. Taste is the capacity to evaluate those outputs against a standard that does not yet exist in the data. AI can produce technically correct code at remarkable speed, but it cannot decide whether that code deserves to exist in its current form. It cannot ask "does this feel right for a person in this context at this moment?" That judgment remains irreducibly human.

How do you develop taste as an engineer?

Taste develops through deliberate exposure to great software, critical analysis of design decisions, and practice shipping to real users. Study products like Linear, Stripe, and Figma as a student, not just a user. Ask why every interaction feels the way it does. Ship your own work and observe how people respond. Build the habit of iterating past "good enough."

What is the ROI of investing in craft?

Companies investing in craft show measurably higher retention (PostHog reports 62% 90-day retention versus a 35-40% industry median), higher revenue per employee (Linear at $1.25M versus Atlassian at $485K), and stronger hiring brands. The upfront cost is real: features take longer to ship. The compounding return is also real: users stay, evangelists emerge, and competitive moats deepen.

Related reading

  • What Is a Product Engineer? - The foundational definition of the role that makes craft possible
  • Building in a World of Slop - How quality standards survive when generation becomes free
  • Product Engineering Culture - The organizational structures that enable taste at scale
  • How to Become a Product Engineer - The career path toward craft-focused engineering
  • Product Engineer vs Software Engineer - Why the distinction matters more than ever in 2026
FB
Felipe Barreiros

Sr. Product Engineer @ AWS

Leading a tech product at AWS with 35 engineers impacting 6.1M customers across 16 languages. 2x founder with exits (acquired by NASDAQ:XP). Coached 12,000 tech graduates. TEDx Speaker. Global Shaper by World Economic Forum. Building product.engineer because 2026 is the year engineers own the full product cycle.

LinkedInX.comGitHubInstagram

Related posts

engineering

Developer Experience AI: Designing DX for Agents and Humans

Developer experience AI agents demand new tooling, testing, and feedback loops. Learn how product engineers design DX when agents are part of the team.

Aug 22 · 18 min read
engineering

Don't Build Slop: 4 Levels of AI Agent Maturity

AI agent maturity spans four levels from copy-paste to autonomous. Learn how product engineers maintain quality at each stage.

Aug 21 · 19 min read
engineering

Dispatch from the Future: What AI-Native Companies Look Like

What an AI-native company actually looks like in 2026. 5-person teams doing what 50 did, built from day one with agents as first-class teammates.

Aug 20 · 18 min read
product.engineer

Owning the whole loop, from idea to impact.

Learn

  • Blog
  • Manifesto
  • Authors
  • RSS Feed

Tools

  • Loops
  • Playbook
  • Discovery
  • Cloud Maturity
  • 5 Whys

Opportunities

  • Jobs
  • Hot Jobs
  • Companies
  • The Role
© 2026 product.engineer
||