Let’s start where most people don’t: privacy.
We keep talking about magical use-cases; meanwhile, data is being hoovered up at industrial scale to feed models that win on volume, variety, and recency.
Regulations are still catching up, incentives are enormous, and inference fills the gaps even when you never “hand over” anything yourself.
The uncomfortable reality:
Privacy won’t vanish overnight, but it will erode in practice — through contact graphs, location exhaust, behavioural patterns, and the simple maths of correlation.
If you’re a leader, treat privacy as an engineering discipline, not a policy PDF.
Build data minimisation into design, use structured retention, adopt confidential computing where it fits, insist on auditable lineage, and prove — not just publish — your controls.
The Killing Season for AI Pilots Is About to Begin
2023–2025 rewarded experimentation.
2026 rewards evidence.
The mood in boardrooms is shifting from:
“What can AI do?”
to:
“What did it deliver?”
Pilots that were loosely attached to outcomes will be shut down.
Capabilities wired into revenue, margin, cycle time, and risk will scale.
That’s not cynicism — it’s the normal transition from theatre to throughput.
How to Survive the Cull
- Move from proof-of-concept to proof-of-value.
- Give every initiative a named owner, a baseline, and two measures: one efficiency, one effectiveness.
- Design for production from day one: data contracts, observability, fallbacks, human-in-the-loop, and clear rollback.
- Impose a 90–120-day runway to show traction; if it can’t clear the bar, stop funding it and redeploy the team.
AI-Driven Layoffs Will Accelerate
People costs remain the largest line on most P&Ls.
When automation reliably compresses task time and error rates, CFOs rebalance.
Some staff will be redeployed.
Many will not.
Remember the lag:
Enterprise programmes launched in late 2022 began paying off through 2025.
2026 is when the harvest scales.
This isn’t moustache-twirling; it’s arithmetic.
If a workflow sees a 30–50% efficiency lift, headcount follows unless you have explicit growth to soak it up.
What Individuals Can Do About It
- Upskill into higher-leverage work (decisioning, orchestration, architecture, governance).
- Become AI-empowered in your current role (workflow design, prompt chains, agent supervision, measurement).
- Reskill into adjacent domains that are slower to automate.
Doing nothing isn’t a strategy.
Why These Three Truths Land This Year
Data Hunger Meets Weak Guardrails
Models that win are fed relentlessly; the commercial incentives to collect and infer will outpace enforcement for a while yet.
The P&L Moment
Boards tolerated theatre during capability-building.
Now they want unit economics, not demos.
Capability Consolidation
Recent analyses show adoption up and value clustering around organisations that fixed data foundations, risk, and integration.
In other words:
the tech isn’t the blocker; operating models are.
What Leaders Should Do This Quarter
1. Set Three Running Programmes with Live Dashboards
- cost take-out
- throughput uplift
- risk reduction
Each needs:
- an owner
- a baseline
- a monthly read-out
2. Rationalise the Portfolio
Keep 10 that scale.
Kill 40 that don’t.
It’s kinder — and cheaper — than letting them smoulder.
3. Rebuild Privacy as a Platform Capability
- data inventory
- minimisation
- retention
- lineage
- confidential compute
- DLP
- policy-as-code
…with evidence you can show auditors.
4. Invest in Capability at the Edge
Train the doers.
My recent piece on serious AI training options for 2026 is written for non-specialists and specialists alike; it’s the most reliable hedge against the labour shift.
5. Align Incentives
Reward:
- knowledge-sharing
- cross-functional delivery
- measurable outcomes
Not slideware.
Bottom Line
2026 separates futurists from executioners.
Focus on:
- problems worth solving
- data you can trust
- outcomes you can measure
Everything else is noise.
