Personal AI took longer than I thought (and arrived sooner)
The category I've been writing about for two years arrived faster than the broader industry conversation acknowledges and slower than the early enthusiasts expected. The shape of what's actually here is more nuanced than either timing claim alone suggests.
End-of-year reflection on the personal-AI category I've been writing about since 2023. The shape that emerged through 2025 has the rare property of being both ahead of and behind expectations, depending on whose expectations you're measuring against. Worth being explicit about both, because the timing claims that get made in isolation tend to oversimplify the actual state.
For context on why I write about this category at all: I've been peripherally circling it for about a decade. In my VMware days I flew to SF corporate often, and the route from the airport always went past the OpenAI building. For a while I told myself I should go check it out. Eventually I did. It was interesting. I struggled to find a clean test case to apply it to the work I was actually doing at the time, and I went back to the work.
So this piece (and the half-dozen before it) is the same instinct revisited with better tools and ten more years of context. The thing I'm circling now is whether the personal-AI category really did arrive, and on what timeline. Both halves of that question turn out to have an honest answer.
Took longer than I thought
The places where personal AI didn't arrive on the schedule the optimistic 2023 read (including mine) would have predicted:
Consumer-friendly principled-personal-AI products. The product layer for "casual user gets the privacy-respecting always-on AI assistant without becoming a platform engineer" still doesn't exist as a credible category. Apple is the most-likely vendor; the bridge product hasn't shipped. By this point in the timeline I sketched, I'd expected it to be a real consumer category.
Data portability and ownership. The legal and technical framework for moving personal AI context between vendors is still patchwork. The MCP momentum addresses tools; the personal-data-and-memory portability remains vendor-specific. The lock-in is real and continues.
Mass adoption of local-first patterns. The principled-user population is meaningfully larger than it was two years ago and remains a small fraction of the overall AI-using population. Most people's personal-AI experience in 2025 is hosted, with the privacy compromises that implies.
The cost-economics shift to favor local. I expected the price compression on local hardware plus the operational maturity of local-AI tooling to make local the default for individuals by year-end. The break-even moved in the right direction; it didn't cross the threshold for most users.
Tooling of personal-AI third-party developers. I expected meaningful developer activity around the principled-personal-AI foundation. There's some; it's smaller than I expected. The cloud-API developer ecosystem absorbed most of the developer attention.
These are real misses on the timing. Two years isn't enough for these specific things to mature; my 2023 framing had an aggressive timeline that didn't survive.
Arrived sooner
The places where personal AI is meaningfully more present than the broader industry conversation acknowledges:
The foundation is genuinely there. Apple Silicon plus open-weights plus mature local-inference tooling delivers a workable personal-AI foundation today. Anyone in the principled-user population can build a daily-use personal AI for the cost of a Mac Studio plus some weekend setup time. Two years ago this was speculative.
The principled practitioner population is real. Small but committed. The community building this stuff out (me, the people I write to, the practitioners I read) is large enough to sustain the open-source tooling, the model release cadence, the practitioner-conversation mass. It's not mainstream; it doesn't need to be to be real.
The hosted-AI assistants accumulated real personal-context capability. ChatGPT memories, Claude's memory features, the various platform-specific personal-context tools, these moved from non-existent to functional through 2025. The privacy story is compromised compared to local; the personal-context-as-product capability is delivered.
Apple Intelligence shipped. Imperfectly, slowly, with the headline features still missing. The foundation is shipped. The pieces are in place; the consumer experience is partial. But the OS-level commitment to on-device personal AI is real and the trajectory is forward.
Local-AI fluency is increasingly common in technical practitioners. The community of people who can set up Ollama, run a local model, build a small personal-AI workflow grew meaningfully. The skill that was niche in 2023 is journeyman in late 2025.
These are real arrivals. The category is more present than the keynote conversation reflects.
What the both-at-once shape suggests
The simultaneous "took longer" and "arrived sooner" pattern says something about how categories form:
The foundation matures faster than the products. Hardware, models, frameworks, primitives, these all moved on the early-2023 timeline. The user-facing products that turn the foundation into mass-market value moved much slower. This is a normal pattern; I underestimated how slow the product layer would be.
The practitioner population grows before the consumer population. The people who could make use of the foundation without product polish materialized first. The mass population that needed the polish came along behind them. The principled-user / casual-user split I wrote about in August is the working-out of this dynamic.
The category is real before it's widely visible. The marketing-narrative version of "personal AI" doesn't yet match the practitioner reality. The practitioner reality is bigger than the narrative because the narrative needs a flagship consumer product to anchor on, and the flagship hasn't shipped.
The eventually-mainstream version may take another 2-3 years. The foundation is ready. The products take longer. The product-and-distribution cycle that turns a foundation into a mass-market category is multi-year work; the principled-user phase is the precursor, not the destination.
What this means for late 2026 and beyond
A few takes for the next two years that fall out of this:
The bridge product gets built, by Apple or by someone else. The foundation-readiness combined with the unmet demand creates the conditions for a flagship consumer-friendly principled-personal-AI product. Apple is best positioned; if they don't ship it, someone else will. By end of 2026 this should be visible.
The principled-user community grows but stays a minority. The pattern I wrote about in August continues. Larger than it was; smaller than the overall AI-using population.
The hosted-AI personal-context capabilities improve faster than the privacy story does. Vendors will keep adding personal-context features; the privacy architecture for them will continue to compromise. The casual user accepts the trade-off; the principled user opts out.
Data-portability standards form, slowly. The MCP-style work for personal AI data and memory will happen. Probably starts as community-driven (a few open-source projects) before any commercial standard emerges. Multi-year horizon.
Apple Silicon's lead on personal AI extends. Continued ANE improvements, continued MLX maturity, continued open-weights friendliness. The lead is structural; it widens before it narrows.
These are takes for the next checkpoint. Worth coming back to in late 2026 to grade.
The honest summary
Personal AI in late 2025 is in the awkward middle phase, the foundation is built, the practitioner reality is real, the consumer-product reality is missing, the mass-market story is incomplete. The "took longer" framing is right for the consumer-product side. The "arrived sooner" framing is right for the practitioner-foundation side. Both are true simultaneously.
The category is the durable one I called in 2023. The shape it's taking isn't quite the shape I sketched. The trajectory is toward what I described; the timeline is longer than the original framing implied. I'm patient about it and still building for it.
The next two years are mostly about closing the bridge between the foundation and the consumer product. The work that gets done in that bridge determines whether personal AI becomes the durable consumer category I bet on or stays a niche the principled users inhabit. I'll be watching closely and staying engaged.
Personal AI took longer than I thought. It also arrived sooner than the keynote conversation acknowledges. Both true at once, and the synthesis is the category as it actually is.