2025 Recap: Grading My Predictions

At the start of 2025, I published my predictions for the year. Now that the dust has settled, let's see how I did.

Figma loses ground to AI coding and prototyping tools

Product designers and PMs adopt AI coding and prototyping solutions more and more. Not the norm yet, but the differentiator.

This one played out largely as predicted, and the uptake towards the end of the year was even more surprising than I expected. Claude Code exploded in adoption, arguably surpassing Cursor as the tool designers and engineers were most excited about. Cursor continued to grow as well, and together they've made AI-assisted coding feel inevitable rather than experimental.

Figma, to its credit, continued to invest in improving their MCP and deepening integration with agentic coding tools, validating my prediction that they'd need to meet developers where they are. But the real shift was cultural: power designers and PMs started building prototypes and even shipping features directly, using AI coding tools as their medium. Interestingly, the "design engineer" identity I mentioned is almost fading away. Not because it failed, but because design and engineering are merging so closely together that the gap the title was meant to fill is narrowing. When designers can code and engineers can prototype, the distinction matters less.

This is exactly what I predicted: not the norm, but the differentiator. Top-performing designers and PD organizations are adopting agentic coding as part of the design process. But scaling it and adopting it more formally into processes, especially at larger organizations, is still slow on the uptake, which tracks with my expectation that we weren't there yet.

🟢 Right

No monolithic AI UI paradigm — apps ship multiple AI interfaces

More apps adopt multiple AI interaction patterns (inline editing, chat, agentic). Browser/OS-level AI doesn't cross the chasm yet. More contextual command interfaces in consumer apps, more canvas-based interfaces in enterprise.

I was partially right here, but the landscape converged more than I expected. Rather than a proliferation of diverse AI UIs within apps, chat-based interfaces à la ChatGPT became the dominant pattern. The innovation has been less about the interface itself and more about what you can plug into it: skills, plugins, data sources, MCPs. The chat window became extensible rather than being replaced.

One thing I didn't predict at all was the rise of the CLI as an AI interface. Tools like Claude Code gained serious adoption, proving that for power users, a terminal-first experience can be more effective than any GUI.

Where I was right: browser and OS-level AI moved much slower than most people expected. Apple Intelligence remained underwhelming, and it took a third-party project like Clawdbot to show us what might actually be possible for AI at the browser level in the not-so-distant future.

🟡 Partially right

Greater crypto adoption via memecoins, easier on-ramps, and DeFi maturation

Memecoin mania drives demand alongside stable yields. Easier on-ramps, chain abstraction, and crypto payment products bring normies onboard.

I was wrong on this one. We didn't see the meaningful worldwide crypto adoption I predicted. Memecoins had their moments but didn't drive the sustained wave of new users I expected. The on-ramp and DeFi maturation story didn't play out in a way that moved the needle for mainstream adoption.

What I didn't see coming was that the most notable growth in crypto adoption came from prediction markets — Polymarket in particular became a cultural phenomenon during the election cycle and beyond. That was the on-ramp, not payments or yields.

🔴 Wrong

Apple buys Snap, future of AR is AI

Apple acquires Snap to bolster AR after Vision Pro underperforms. AI assistance makes Watch and AirPods the primary mobile AI interfaces.

The bold call here was wrong. Apple did not buy Snap. But they did make a significant acquisition in the human-computer interface space: Q.ai, for nearly $2 billion, a company focused on AI that can recognize whispers in noisy environments and detect subtle facial muscle movements. It's clearly aimed at next-generation wearable interaction, which rhymes with my broader thesis about Watch and AirPods becoming primary AI surfaces.

Meanwhile, it was OpenAI, not Apple, that made the splashier hardware move, acquiring Jony Ive's io for roughly $6.5 billion to build a new consumer device. Reports indicate it will be audio-first, with Ive framing it as a chance to reduce device addiction and reimagine what it means to interact with a computer. That's close to my prediction about audio as the primary interface for mobile AI. It's just coming from a different company, and it's still early.

🔴🟡 Wrong on the main call, directionally interesting

Traditional media falls, new influencers rise across a fragmented landscape

Social channels get jammed with AI slop, users retreat to trusted personalities. Influencers gain power, niche creators focus on community and paid subscriptions.

This one played out almost exactly as predicted. "Slop" was named 2025 Word of the Year by Merriam-Webster, and usage of the term grew 9x compared to 2024. Estimates suggest anywhere from 20 to 90 percent of surfaced feed content is now AI-generated depending on the platform — Pinterest, YouTube, and Facebook were hit especially hard.

To be fair, the shift toward creators and away from traditional media isn't solely because of slop. The amount of AI-generated content is definitely increasing and it's palpable — the quality of incumbent social media channels does anecdotally seem to be decreasing overall. But it hasn't gotten so bad that it's caused a mass mainstream exodus just yet. The broader fragmentation trend was already well underway. Social and video platforms overtook TV and news websites as primary news sources in the US for the first time, and that consumption increasingly funneled through individual creators rather than institutional outlets. Podcasters, YouTubers, and Substackers continued to eat into traditional media's audience, and micro-influencers gained outsized power in niche communities.

The paid subscription and community-building trend also held. As feeds got noisier, audiences showed a willingness to pay for signal, whether through Substack, Patreon, or private communities. The desire for authenticity and human connection became the counterweight to slop.

🟢 Right

Captain America and Thunderbolts tank, Superman underperforms, Mission Impossible is #1

The cultural repudiation of the comic book formula continues. Mission Impossible tops the box office as a testament to real-stakes spectacle.

I was right that Captain America and Thunderbolts would underperform. Captain America: Brave New World landed at #11 domestically with $200M, and Thunderbolts* at #13 with $190M. Both disappointing by MCU standards. The Marvel fatigue I predicted continued to play out.

Where I was wrong: Superman didn't just do well, it was the #3 domestic film of 2025 at $354M and $619M worldwide. I predicted it would underperform despite a fresher take from James Gunn. Instead, that fresh take resonated far more than I expected. The appetite for hopeful, well-crafted superhero films wasn't gone; audiences were just tired of the MCU formula specifically.

Mission Impossible: The Final Reckoning came in at #12 domestically with $197M on a reported $400M+ budget. Far from #1. It did better internationally ($599M worldwide), but this was a miss. The top of the 2025 box office belonged to A Minecraft Movie ($424M) and Lilo & Stitch ($424M). The year was dominated by family-friendly IP, not the real-stakes spectacle I bet on.

🟡 Mixed. Right on Marvel fatigue, wrong on Superman and Mission Impossible

Overall Score

3 / 6

Beyond the predictions themselves, what surprised me most about 2025 had little to do with what I was tracking at the start of the year.

ChatGPT became overwhelmingly the consumer AI interface of choice, but there isn't a lot of evidence in the mainstream that people are using it for much more than a glorified Google. The deeper, more transformative adoption happened in the workplace, and specifically with Claude. Anthropic's ability to iterate and innovate for different work use cases (coding, writing, research, analysis) was genuinely surprising. Claude went from an alternative to the default for a growing number of professionals and teams.

What caught me most off guard, though, were the already seemingly large shifts in the tech labor market. The speed at which AI reshaped hiring expectations, team structures, and what it means to be productive as an individual contributor moved faster than I think most people, myself included, anticipated heading into the year.