Most AI commentary is opinion in a nice jacket. I wanted something you could check.
So this year I started Alchemy Research: one original report per quarter: each answering a key question leaders struggle to get straight answers to; on the size of the UK AI Market, the state of AI Adoption, the AI sentiment gap, and the AI safety ratings of the major labs. Each lives on a page I keep updated annually.
In addition, we have the "AI Black Box" - a digest of AI harms, which is updated quarterly on a slightly offset cadence. Why? Well because when the headlines vanish, the lessons often disappear with them.
As we approach the end of the year, I wanted to look back over the first year of our original research - and give a brief first edition summary of each. Numbered in the order we published them:
Alchemy Research β π¦ Q1 β π© Q2 β π¨ Q3 β π₯ Q4 β β¬ Black Box
π¦ How big is the UK AI market? Q1
ββ
The UK tech sector is now worth Β£1.2 trillion. AI is a third of it.
Β£384bn: AI's share of UK tech in 2026. That is 32% of the sector, more than double its share five years ago (Tech Nation).
The numbers behind it:
- Β£8.2bn of UK AI startup funding in the first half of 2026 alone, already ahead of all of 2025.
- The UK AI sector has grown three times faster than France and twice as fast as Germany over three years, and ranks as Europe's number one AI ecosystem.
- Β£1.6bn of government money is behind it: the Β£500m Sovereign AI Fund plus a Β£1.1bn Hardware Plan.
The caveat matters as much as the headline. 30% of UK founders call AI a bubble in the same survey, and some investors say Β£500m is a drop in the ocean. Funding is the input. The open question is the outcome.
Read it here: https://alchemy.consulting/uk-ai-market-size-2026
π© Why does adoption stall? Q2
ββ
Everyone is piloting. Almost nobody has shipped.
88% of organisations use AI regularly. Only 37% report any EBIT impact. Just 6% qualify as AI high performers (McKinsey).
Four 2026 surveys, four different ways of asking the same question, one uncomfortable answer. Between 11% and 44% of organisations are scaling AI enterprise-wide, and the true figure sits nearer the bottom: KPMG puts genuine scale at 11%. MIT NANDA found 95% of GenAI pilots show no P&L impact.
The cause is the organisation, not the model. It is the J-Curve in the wild: the dip that follows adopting a capability faster than you redesign the work around it. The firms pulling ahead are not the ones with the most pilots. They are the ones furthest through the trough.
Read it here: https://alchemy.consulting/ai-adoption-research
π¨ The AI sentiment gap Q3
ββ
If you only ever looked at LinkedIn, you coudl be forgiven for thinking AI is a universal boon for humanity. Spend your time on TikTok and you find the absolute opposite. There is a growing sentiment gap between the people building AI and the people who had to suffer its consequences.
Public net sentiment on AI has fallen from -44.2 to -62.5 in a year (Bentley-Gallup).
The gap shows up everywhere you look:
- Experts vs public: 73% of AI experts expect AI to improve how people do their jobs. Only 23% of the public agree. That is a 50-point gap (Stanford HAI).
- Boardroom vs front line: 81% of CEOs trust AI-driven delivery decisions. Only 50% of QA directors do (Tricentis).
- Gen Z: the share "angry" about AI rose from 22% to 31% in a year, even though about half use it daily or weekly (Gallup).
This is not a PR problem. Public trust now tracks direct product experience more than media coverage, so better messaging cannot close a gap that lived experience is opening. It's not sensationalist to describe this as a crisis of public confidence. And it is the stuff that revolutions are made of!
Read it here: https://alchemy.consulting/ai-sentiment-gap-research
π₯ The AI Safety Gap Q4
ββ
Euro NCAP does not average a car's safety. It caps the star rating at the weakest pillar, so a manufacturer cannot hide a gap behind strengths elsewhere. I applied the same rule to nine AI providers.
Only one of nine providers reaches four stars. Four sit on one.
Microsoft gets four stars. OpenAI, Google and Anthropic get three. Meta gets two. Mistral, xAI, Qwen and DeepSeek get one.
One pillar I scored separately, because every provider fails it: protection for vulnerable users such as minors, applicants and patients. Several providers have age controls, but nobody has independently verified outcomes. That is an industry blind spot, not one firm's problem.
A fair warning on method: this is an evidence-based scorecard built from public disclosures, not a controlled test. Missing evidence counts as a gap to disclose, not proof of failure. The clock matters too. The EU Product Liability Directive lands on 9 December 2026, and brings software and AI into the legal definition of a defective product.
Read it here: https://alchemy.consulting/ai-safety-gap-research
β¬ The AI Black Box Q1 Q2 Q3 Q4
ββ
Every quarter, plenty of AI failures make headlines, then quietly disappear before anyone learns from them. The AI Black Box files them properly: what happened, who it hit, what changed as a result.
Issue 01 covered:
- KPMG pulling a flagship AI report after just 5 of 45 citations checked out.
- EY's fraud report containing fabricated sources.
- A chatbot that invented a medical degree, a licence number and a diagnosis.
- Peer-reviewed Stanford research showing AI bias in hiring.
Free. Quarterly. Sourced. No speculation.
Read it here: https://alchemy.consulting/AI-BLACK-BOX
Why bother?
Because a number you can trace beats an opinion you cannot. Every report links its sources, and every report gets revised when the evidence moves.
Take what is useful. Argue with the rest. Tell me where I am wrong.
Yours,
ππΆππΎπΉ π±πΎπβ―π
Everything above lives in the Insights Archive.