Sketching Is the Last Slow Step Worth Keeping

Every part of design got cheaper and faster in the last three years except the thinking. Sketching is the practice that protects the thinking, which is why it survives.

In 2026 I can prompt Figma Make to spit out a passable first-draft layout in the time it takes me to scribble a rectangle. I can ask Cursor to wire it up in another minute. Every part of the design pipeline that used to be slow is now cheap. So when I notice that I'm still — every week, on every project that matters — sitting down with a pen and a stack of A4, that means something.

What it means: the artifact was never the point. If sketches were a deliverable, AI would have killed them years ago. They're an aid to thinking, and thinking is the part that didn't get cheaper. Anything in your workflow that survived the last three years of tool acceleration is, by selection, doing a job the tools can't do. Sketching is one of the clearest examples.

I almost gave it up twice. Once when Figma's auto-layout got good enough that I could lay out screens faster digitally than on paper. Once when the prompt-to-mockup tools got good enough that I could skip layout altogether. Both times I noticed within a week that my designs got worse — not slower, worse — and went back. That feedback loop is the whole essay.

1. A sketch is a thinking instrument, not a deliverable

This is the load-bearing sentence and most "how to sketch better" advice misses it: the crappiness of a sketch is not a bug to overcome. It is a feature you are paying for. When the thing on the page is obviously rough, your brain treats it as provisional, which is exactly the state you want to be in when you're choosing between three layouts that might all be wrong.

The moment a sketch looks "good," you stop being willing to throw it away. The moment a Figma mock looks like a screen, you start defending it. That defensiveness is poison early in a project. Sketches buy you the right to abandon ideas cheaply, which is the only mechanism by which the good ideas eventually surface.

The implication that surprised me when I noticed it: if you find yourself making your sketches look nicer — neater lines, consistent spacing, fewer scratch-outs — that's a signal you've drifted from thinking into presenting. Different tool, different stage. Switch venues.

2. Divergent and convergent are different jobs. Most tools collapse them.

Design work has two modes and they want completely different conditions. Divergent mode is where you generate alternatives, throw away bad ones, keep going. Convergent mode is where you take one promising direction and polish it into something real. Conflating them is the single most common process failure I see, and 2026's tooling has made it dramatically worse.

The reason is that every AI design tool I've used is convergent by default. You prompt, it gives you one polished thing. You can ask for variants but they're variants on a theme it already committed to. That's fine when you know what you're refining. It's actively harmful when you don't yet know what you're solving. The first thirty minutes of any real design problem belong to divergence, and the cheaper the artifact, the better divergence goes.

Paper sketches are the cheapest artifact I know. That's the only reason I still use them. If something cheaper showed up tomorrow — a stylus interface that produced equally throwaway results — I'd switch in a week. Until then, paper.

3. Paper for the first thirty minutes. Figma for the next thirty hours.

I tell every designer I mentor the same setup: keep a stack of plain A4, two pens (one black, one a contrast color), and a phone with a scan app open at all times. That's the kit. The tablet, the stylus, the dedicated sketching app — none of it has paid off for me the way paper has, because all of it makes the artifact too good.

The workflow is unromantic. Sketch for twenty or thirty minutes, no more. Photograph the page with iPhone or Android's built-in scanner — both produce clean, cropped, contrast-corrected images in a single tap. Drop into FigJam or a Figma file. Stop sketching. Move to digital.

The mistake to avoid is keeping the paper open while you start to build. Paper is a divergent venue and digital is convergent. Trying to do both at once means doing neither well. When you switch venues, switch modes. Close the notebook. Open the file. Different work.

4. FigJam is the right second venue, not the first

Six years ago the second venue was a whiteboard in a room. In 2026 it almost never is, because the team is almost never in the same room. FigJam (or Miro, depending on your org) is now the default surface for shared sketch review, and the pattern that works is paper-first, FigJam-second — not FigJam-first.

The reason: FigJam invites polish. Templates, stickies, shapes that snap. Within five minutes of opening a blank FigJam board, people start building instead of sketching, and the divergent mode is over before it started. Bring photographed paper sketches in and FigJam becomes what it's actually good at — a place to annotate, vote, group, and decide.

The annotation pattern I've settled on: a photo of the sketch in the center of a frame, stickies around the edges, colored dots for vote-style feedback, arrows for sequencing. A FigJam board is a decision-making artifact, not a creative one. Start the creative work elsewhere and bring it in once you have something to decide on.

5. Sketch what AI can't. Skip what it can.

The biggest change to my actual sketching practice in the last two years is that I sketch less surface and more structure. Layouts, type hierarchies, color palettes, icon choices — all the things I used to spend sketch time on — I now skip and hand to Figma Make or Claude with a one-line prompt the moment I know what screen I'm building. The AI is faster than my pen at all of them and the variance is acceptable.

What I still sketch by hand: decision points in a flow. Alternative paths a user could take. The handful of moments where the system has to behave one way or another and I haven't yet decided which. The shape of an interaction across multiple screens, where the connections matter more than the contents. Anything where the question is "which of these should we do" rather than "what should this look like."

The division of labor that's emerged: I sketch the bones, the AI fleshes them out. The hour I save on pixel-mockup work goes back into the bones, which is where the leverage actually is. The teams I see struggling with AI-assisted design are usually the ones who've kept hand-mocking the easy parts and short-changing the bones.

6. Wireflows are where team alignment happens or doesn't

A wireflow — the sequence of screens with the connections between them — is the most useful sketching artifact I produce. More useful than any individual screen sketch. The reason is social: a wireflow is the thing the rest of the team can actually argue with. A single screen is too local to argue about productively; a polished prototype is too finished to argue with comfortably. The wireflow sits exactly in the middle.

The shapes I bother with are minimal. Boxes for screens, arrows for transitions, sticky notes or margin annotations for decision points and edge cases. The whole thing fits on one page or one FigJam frame. If your wireflow doesn't fit in a single view, you're documenting, not aligning. Documentation is a different artifact for a different purpose. Don't conflate them.

The questions a wireflow has to answer to be useful: where does the user start, what happens at the moment of decision, what does failure look like, and where do they land. If your wireflow doesn't surface those four, polish later. Argue about those four first.

7. Accessibility is a sketch-stage decision

WCAG 2.2 added two criteria that are trivially cheap to honor at sketch time and brutally expensive to retrofit. Criterion 2.5.8 sets a minimum interactive target size of 24×24 CSS pixels. Criterion 2.4.13 sets requirements for visible focus indicators. Both are layout constraints. Both belong on the page next to the layout, not in a follow-up audit ticket three sprints later.

The annotation that's saved me the most pain is the smallest interactive target on the sketch, drawn to scale, with a tiny "24" written next to it. Once it's on the page, the whole layout has to make room for it, and every reviewer who sees the sketch internalizes the constraint without anyone having to lecture them about accessibility. The constraint becomes ambient instead of contentious.

The trap to avoid is treating accessibility as a phase. It's not a phase. It's a constraint, and constraints have to be present when the decisions are being made. Layout decisions made without accessibility constraints in the room are layout decisions you'll be remaking, expensively, in code, six months from now.

8. AI is great at refining a sketch. Terrible at having one.

The most common mistake I watch designers make in 2026 is starting a screen in a prompt. They open Figma Make or Claude or whatever, type "design a settings page for…" and let the tool produce a first draft, which they then react to. This feels faster. It is the slowest possible way to design, because it locks in framing decisions you never made consciously.

What the tool produces is its best guess at what an average settings page should look like, given an average framing of the problem. If your problem isn't average — and the problems worth designing rarely are — you've just anchored yourself to the wrong shape, and every subsequent iteration is working against that anchor.

The order that works: sketch the bones first, in your hand, in five minutes. Photograph it. Prompt the AI with the sketch as a reference. Now the tool has constraints it can refine against, rather than a vacuum it has to fill. The sketch is what tells the AI what problem you're solving. Skipping it is skipping the most important step.

9. Practice that compounds, and the kind that doesn't

The advice I used to give junior designers about sketching practice was to draw a lot — copy real apps, fill notebooks, build muscle. It's still right, but it's incomplete. The practice that compounds in 2026 is the practice that builds your library of decisions, not your library of shapes. Shapes are now free. Decisions are still expensive.

The exercise I actually recommend: take a screen you use every day — your email client, your calendar, your IDE — and sketch three alternative layouts in fifteen minutes. Not better layouts. Just different ones, with notes about which trade-off each one is making. The point is to feel, in your hand, the cost of each decision and the alternatives it forecloses. That feeling is what AI tools can't give you and what good design depends on.

The deeper practice is to sit with the discomfort of an unfinished sketch. The convergent tools are seductive precisely because they relieve that discomfort instantly. Learning to stay in the divergent mode for an extra fifteen minutes — to keep generating when your brain wants to start polishing — is the single skill that most separates the designers I want to work with from the ones I don't.

The thread

Every tool change of the last three years has cheapened the artifacts of design — the mockups, the prototypes, the polished frames — without cheapening the thinking. Sketching is the practice that protects the thinking. As long as the artifact-production half of the pipeline keeps getting cheaper and faster, the value of the thinking half goes up, not down. Sketching is the most direct way I know to spend time in the half that's appreciating.

If you take only one thing from this and try it this week: tomorrow morning, before you open any design tool, spend twenty minutes with paper on whatever problem is in front of you. Photograph what you produce. Then open your tools. Notice what the constraint of having sketched first does to the work that follows. The answer is the whole point.