Futurism: where the world might be in ten years

Not a flying-car post#

Futurism usually dies in one of two ways. Either it promises jetpacks and gets laughed at, or it extrapolates last quarter's demo into a civilisation rewrite and gets believed for eighteen months.

I am more interested in the third option: what already works in pockets, what the incentives reward, and what still has to survive contact with regulators, attackers, and ordinary people who did not ask to be early adopters.

So here is a 2036 sketch from late 2026. Not a forecast I will bet the house on. A set of bets about direction, with the same honesty I use when a customer asks whether a security product will "solve identity."

AI becomes boring infrastructure#

By 2036, arguing about whether AI is "real" will sound like arguing about whether the cloud is real in 2014.

The interesting shift is not a single god-model that replaces every job. It is that model inference becomes a utility layer under almost every software surface: search, CRM, support, code review, document drafting, fraud scoring, medical triage, and the half of knowledge work that was always "read this, summarise that, draft a reply." Most people will not open a chat window to "use AI." They will just notice that software finally does the tedious middle of the task.

That creates a second-order effect people underweight. Organisations that treat models as a productivity feature will lose to organisations that treat them as an operating system for work: versioned prompts, evaluation harnesses, human approval gates, audit logs, and clear ownership when the model is wrong. The winners will look boring. The losers will have a GenAI slide and a data leak.

I also expect the cost curve to keep doing what compute curves do. Capability that felt magical in 2025 becomes table stakes. Differentiation moves upstream into proprietary data, workflow integration, and trust. Same story as every previous platform shift.

Trust gets harder on purpose#

If you have read my post on AI-based attacks, you already know the direction.

By 2036, a familiar face on a video call and a familiar voice on the phone are evidence of media, not evidence of identity. Deepfakes do not need to be perfect. They need to be good enough to beat the verification habits companies still run on "I recognised them."

The response that survives is process, not better eyes:

• Dual control on money movement that a call cannot override
• Callbacks to known-good numbers
• Shared secrets and challenge questions
• Hiring identity checks that assume the Zoom face can lie
• Continuous authentication that cares more about behaviour and device posture than a one-time selfie

Cybersecurity does not get easier. Attackers get a factory for social engineering. Defenders who still equate recognition with authentication will keep paying tuition.

Autonomy arrives unevenly#

Physical AI is the part of the Tesla argument that still matters even if you do not care about the stock. Robotaxi, warehouse robots, and humanoids do not need to conquer every job to reshape labour markets. They need to win the dull, dangerous, repetitive slice in places that can afford the fleet and the liability stack.

My 2036 picture:

• Unsupervised ride-hailing is normal in a handful of US and Chinese metros, and still politically radioactive in much of Europe
• Factories and warehouses run more humanoid and specialised robots than press releases promised in 2025, and fewer than the keynotes implied
• Most care work, trades, and edge cases still need people, because edge cases are where reality lives
• The economic story is utilisation and unit cost, not "robots replaced Tuesday"

If you want the longer version of why sequence beats calendar hype, I already wrote it in the Tesla master plans piece. The ten-year lesson is the same: timing slips; direction compounds.

Energy is the quiet constraint#

Every AI datacentre, every EV fleet, every heat pump rollout eventually collides with the same boring question: where do the electrons come from, and can the grid hold them?

By 2036 I expect more solar, more storage, more local generation, and more political fights about permitting and transmission than about whether batteries "work." Energy storage stops being a Tesla footnote and becomes ordinary infrastructure, the way broadband did. That does not mean climate is solved. It means the buildout is visible enough that people argue about siting and cost instead of physics.

Nations that can generate, transmit, and cool compute will have a strategic advantage that looks a lot like industrial policy with better cooling towers.

Healthspan, not immortality#

I am not going to pretend we all live to 140 with perfect skin.

What I do expect is that the conversation moves from "anti-ageing as fringe" to "healthspan as a normal adult project," the same way training and bloodwork already crept into mainstream fitness culture. Peptides, metabolic drugs, continuous monitoring, and personalised protocols will be more common, more regulated, and more unevenly distributed. Some of that will be evidence-based. Some of it will be expensive noise. The peptide boom is an early chapter, not the last one.

The practical outcome for most people is not sci-fi medicine. It is earlier detection, better metabolic control, and a wider gap between people who treat their body as a system they maintain and people who wait for a crisis.

Work: fewer blank pages, more judgement#

Remote and hybrid do not reverse. Tools get better at async collaboration and worse at pretending presence equals output.

The job market polarises around judgement. If your role was mostly assembling known answers into a document, models eat the middle. If your role is deciding what matters, owning risk, negotiating reality with other humans, or keeping production systems honest, you get more leverage, not less. Sales engineering, security architecture, and infrastructure ops do not vanish. They get a co-pilot and a higher expectation that you can still explain the decision when the co-pilot is wrong.

Education lags, as usual. Credentials that prove you can prompt a model will be less valuable than proof you can ship, verify, and take responsibility.

What will not change#

This is the part futurism posts skip because it is not exciting.

People will still want status, belonging, and someone to blame. Scams will still work on urgency. Governments will still regulate after the damage is visible. Homelabs will still be more fun than enterprise change boards. French bulldogs will still have opinions.

And the organisations that win will still be the ones that can operate: clear ownership, boring controls, restore drills that actually restore, identity that assumes compromise, and leaders who prefer a working system over a visionary slide.

A usable 2036 checklist#

If you want something more actionable than vibes:

1. Treat AI as infrastructure. Build evaluation, logging, and human gates now.
2. Retire "I saw them on the call" as authentication.
3. Bet on electrification and compute, but plan for grid and cooling constraints.
4. Invest in healthspan habits that compound whether or not the miracle drug arrives.
5. Build skills that survive automation: judgement, communication, systems thinking, and the discipline to verify.

Bottom line#

In ten years the world will not feel like a different planet. It will feel like this one, with model inference under the glass, harder identity problems, robots in the places economics allow, denser energy fights, and a wider gap between people who adapt their operating system and people who wait for a press release to tell them the future arrived.

The useful futurism question is not "what magic shows up." It is "which defaults become normal, and who is ready when they do."