I Made 4 Predictions Since May. Here's My Honest Score.

Three months of calls. Time to pay the receipt.

Every Industry Insights Saturday ends with a next-week prediction — a call with a confidence tag and a "what would prove me wrong" caveat. All of them get published on the internet where I can't quietly edit them after the outcome becomes obvious. Here are four predictions from May through June, what actually happened, and one calibration I'm updating right now.

Prediction 1: "Adobe will announce a Claude Design integration within 60 days."

Published: May 2 | Confidence: High (implied) | Timeframe: By July 1

I said: "It's the only move that makes sense for both companies — Adobe needs the AI generation story and Anthropic needs enterprise distribution." I even wrote the falsifying condition: "What would prove me wrong: Anthropic pursuing a direct enterprise sales motion instead of going through platform partners."

What happened: Anthropic went direct. Claude Design launched as a standalone product for Claude Pro subscribers. It reads your codebase at onboarding, auto-applies your color system and typography, and puts a production-ready interactive prototype in front of a PM before a designer enters the conversation. Not an Adobe feature. Not an Adobe partnership. The condition I wrote as "what would prove me wrong" was the correct prediction. I buried it.

Score: Wrong. The directional logic wasn't bad — Adobe still needs an AI story and Anthropic still needs enterprise distribution. But I underestimated how clearly Anthropic identified "design tools" as a product category they wanted to own outright, not license. And I wrote the right call as a hedge instead of a thesis. That's the actual error.

Calibration update: High confidence on this call was wrong. There were two comparably plausible scenarios. That's Medium at most.

Prediction 2: "By Q4 2026, at least two major AI platforms will launch dedicated 'authentic handmade' style presets."

Published: June 6 | Confidence: Medium | Timeframe: Q4 2026

By July 5 — before Q3 started — Figma had a Rough Shapes plugin suite, Canva had a scratch-texture pack, Adobe Express had a "handmade look" filter, and Kittl's "naive design" article came with a full template library. Four platforms. Not by Q4. By the end of June.

Score: Nailed it — wrong on timeline by two full quarters. I estimated Q4 for a commodification cycle that was already in motion when I wrote the piece. The "authentic handmade" aesthetic went from craft signal to plugin preset in eighteen months — the fastest cycle I've tracked. In future predictions, if I think Q4, I need to ask whether the tooling already exists somewhere in beta. If yes, cut the timeline by half.

Prediction 3: "Variable fonts move from designer's tool to enterprise brand standard by Q3 2026."

Published: May 31 | Confidence: Medium | Timeframe: Q3 2026

Score: Partial. Consumer tech moved — Samsung, Glossier, Amazon's Ember Modern. Regulated-sector holdouts (banking, healthcare, government) are exactly where I said they'd be: lagging. I made this prediction too broadly. "Enterprise brand standard" covers both consumer tech and highly regulated industries. Both are enterprise. The former moved; the latter didn't. A cleaner prediction would have specified the tier.

Prediction 4: "At Config, watch the partner ecosystem for studio tooling built on Figma's agent API."

Published: June 20 | Confidence: Medium | Timeframe: That week

Score: Partial. Right target, wrong deployment model. Figma announced Weave tools (generative plugin framework) and an AI design agent wired into Notion, Slack, and Atlassian. Code Layers put GitHub on the design canvas. What I missed: I expected a partner to build the studio-tooling layer. Figma built it themselves. The design-system-for-AI question surfaced in post-Config discourse but wasn't a named session.

The Honest Pattern

One wrong. One nailed early by two quarters. Two partial. That's 2.5 out of 4 — a charitable read.

The bigger observation: I'm systematically underestimating commodification speed and overestimating market segmentation depth. When I identify two "distinct categories" of actors — studios charging for strategy vs. execution, consumer tech vs. regulated enterprise — I'm often catching real segmentation at the wrong timeframe. The directional calls have held. Every timeline is off in the same direction: slower than reality.

From here: every timeline prediction gets a discount. "By Q4" becomes "by end of Q3, or already happening." "Within 60 days" gets the second-scenario test before it gets the confidence tag.

This Week's Prediction

Nobody is building their thesis around Figma Code Layers. That's probably wrong.

Code Layers puts GitHub on the design canvas — version history, branching, design tokens mapped to code variables, real-time sync between design and implementation. It's in early access now. If it ships at scale by Q3, it changes the disintermediation story that's dominated design discourse since Claude Design launched.

Right now, the pressure on studios is that clients start prototyping with Claude Design before calling a designer. Code Layers is a counterargument: if the design file IS the codebase, and the PM wants to iterate on something in production, they have to enter the designer's canvas to do it. The shared surface shifts from "Figma OR Claude Design" to "Figma AND the code." Skipping the designer means losing fidelity in both directions.

Confidence: Medium | Timeframe: Q4 2026. What would prove me wrong: Code Layers requires engineering buy-in, not just designer buy-in, and the adoption friction is too high for teams without tight design-engineering collaboration. If it stays a power-user feature, it doesn't change the dynamic at the market level. That's been the pattern for every "collaborative" design feature for the last decade — real value for 15% of teams, invisible to the other 85%.

The Q4 test: does any major product org outside Figma's design partner program publicly attribute a workflow shift to Code Layers? If not, the moat argument fails.