AI Horizon Map

Assumptions

Reasoning, memory, tool use and self-improvement keep compounding until frontier systems can transfer robustly across most cognitive domains at roughly human level.

Blockers

The last 20% of reliability, autonomy and world-model grounding proves far harder than scaling advocates expect.

Implication

Build for an agent-native world now: outcome-based products, tiny oversight teams, and businesses that assume intelligence becomes abundant before trust does.

Assumptions

Tool use, memory, permissions, evaluation and error recovery improve fast enough that agents can own large volumes of queue-based knowledge work without human approval loops.

Blockers

Identity, auditability, exception handling and liability stay unresolved long enough to stop organisations trusting unattended execution.

Implication

Build narrow, high-volume domain agents now and wrap them in approval tiers, rollback paths and outcome-level monitoring.

Assumptions

Vision-language-action models, dexterity, battery performance and manufacturing scale improve together quickly enough to make general-purpose robots commercially routine and cheap.

Blockers

Reliability in cluttered environments, safety certification and service economics fail to move from demo quality to fleet quality.

Implication

Start designing robot-ready workflows, facilities and software now, because the integration layer will matter as much as the hardware.

Assumptions

Long-horizon coding agents become strong enough at planning, editing, testing, migration and repository memory that humans mostly specify, review and steer.

Blockers

Security, reproducibility, architecture drift and repo-specific context keep agents from owning production changes at scale.

Implication

Rebuild engineering around evaluation, review policy and product judgement, because typing and boilerplate stop being the scarce skill.

Assumptions

AI tutoring becomes dramatically better and cheaper than conventional content delivery, and assessment adapts fast enough to preserve trust in learning outcomes.

Blockers

Credentialing inertia, safeguarding, procurement cycles and political resistance keep institutions tied to legacy delivery models.

Implication

Build assessment, coaching, learning-record and teacher-orchestration products rather than more static content libraries.

The pattern is no longer “open follows years later”; it is “open absorbs frontier ideas fast enough to matter commercially”. Stable Diffusion proved that closed advantages can leak into public ecosystems, LLaMA broke the sense that only a few labs could play, and DeepSeek-R1 showed that open reasoning can arrive with real teeth. If open gets to 90–95% of frontier capability at a fraction of the cost, permanent practical parity is enough to change who wins distribution.

o1 made it plausible that test-time compute is not just extra verbosity but a new way to buy capability on hard tasks. Once models can spend inference budget deliberately, the product question changes from “how fluent is it?” to “how much thinking is this worth?”. That creates a genuinely new optimisation space for models, interfaces and pricing, and it is why reasoning now feels like a separate frontier rather than a mere feature.

Routine, high-frequency AI use wants low latency, predictable cost and privacy by default. Once a model is good enough to summarise, transcribe, draft, organise and personalise locally, shipping every interaction to the cloud starts to look like a design relic rather than a necessity. Cloud will still do the heavy lifting, but local could become the default surface most people touch first.

When a field starts inventing shared protocols, permission layers, eval harnesses and operational controls, that usually means the pattern is becoming real. Agents that touch tools, data and workflows need governable plumbing, and that need does not disappear even if today’s demos are messy. The category may still be early, but the infrastructure demand looks structural rather than faddish.

Vibe coding lowers the hardest barrier for many would-be builders: getting from intent to working software. That means domain experts can finally ship internal tools without having to become full-time engineers, which is how real capability usually spreads inside organisations. If guardrails improve, the organisational effect could look less like chaos and more like the spreadsheet moment for software creation.

The useful definition is economic: a system is AGI when it can reliably perform most paid knowledge work at human level or better, across domains, without task-specific retraining. This framing is measurable, matters commercially and treats “general” as a threshold of breadth rather than a philosophical claim about understanding. Reasoning models plus tool use already point at this definition, even if the threshold is years away.