Technologies I Think Will Matter Over the Next Decade

Lists like this age quickly, and that's fine — the value isn't in being right in every detail years later, it's in making the reasoning explicit enough to revisit and correct. This is a personal, opinionated read on which technology areas look most likely to matter over the next decade, held loosely enough to be wrong about the specifics.

Fact, as a starting point: the technologies that ended up mattering most over the last decade — cloud computing, smartphones' full maturation, and now large language models — shared a pattern: each moved a capability from expensive and specialised to cheap and broadly accessible, and the interesting effects showed up downstream of that access shift rather than in the core technology itself. That pattern is a reasonable lens for thinking about what comes next.

AI agents and autonomous tooling. Not as a single dramatic "AGI" moment, but as a steady expansion of the range of well-defined tasks a system can complete with real autonomy and minimal supervision. Opinion: the more interesting story here is quiet and infrastructural — reliable automation of narrow, well-scoped work — rather than the more dramatic autonomous-general-assistant narrative that dominates the marketing.

Energy storage and grid-scale batteries. Fact: battery costs have fallen sharply over the past decade and continue to, which changes the economics of renewable energy from clean-but-intermittent to clean-and-increasingly-reliable. Opinion: this is under-discussed relative to its likely impact, because it's unglamorous infrastructure rather than a consumer-facing product.

Biotechnology, particularly diagnostics. AI-assisted drug discovery and cheaper, faster diagnostic tools both benefit from the same pattern of falling cost and rising accessibility that reshaped software. Prediction, held loosely: the more immediate, broadly felt impact over the next decade is likely in diagnostics and monitoring becoming cheaper and more accessible, ahead of the slower, more heavily regulated process of new treatments actually reaching patients.

Local and edge computing. As models get more efficient and hardware gets more capable, a meaningful share of AI workloads that currently require a cloud API will run locally on ordinary devices. Analysis: this matters most for privacy-sensitive use cases and for reducing dependency on any single provider, rather than for raw capability, where cloud-hosted frontier models will likely stay ahead for the hardest problems.

Automation of physical, not just knowledge, work. Robotics has lagged software's pace of improvement for a long time, partly because the physical world is a much messier training environment than text. Prediction, held loosely: that gap likely starts narrowing meaningfully within the decade, as techniques developed for language and reasoning models get adapted to physical control, though the timeline here is genuinely more uncertain than for software-only technologies.

What's probably overrated, held just as loosely: fully autonomous, general-purpose humanoid robots as a near-term consumer reality, and blockchain-based systems finding mainstream utility beyond the specific niches — payments in unstable currency environments, certain kinds of provenance tracking — where they've already found genuine, if narrow, product-market fit.

Opinion, overall: the safest bet across all of these isn't any specific technology, it's the pattern underneath them — capability that used to be expensive and specialised keeps becoming cheap and accessible, and the biggest effects keep showing up not in the technology itself but in who gets access to capability they didn't have before, and what they do with it. Betting on that pattern continuing has a better track record than betting on any single technology's specific timeline.