The Reliability Gap Has Closed. Why We're Hiring an AI Knowledge Architect.

Earlier this year I wrote “My 3 AI Predictions for Market Research in 2026”. The underlying thread running through all three was that the next phase of AI in our industry would be defined less by model capability and more by how organisations harness it. A few months on, I hold that view more strongly than when I wrote it.

The reason is simple. The AI of 2026 feels different.

Something has shifted

I won’t pretend the old problems are fully solved. But anyone doing serious work with these tools will recognise what I mean. Outputs from the frontier models are more grounded, more traceable, and more predictable than they were even twelve months ago. Agentic capability; the ability to chain tools, follow multi-step logic, and execute reliably across a workflow; has matured in ways that change what’s actually possible.

This is the difference between AI being an interesting assistant and AI being something you can build serious institutional processes on top of. And for any research or consulting organisation thinking about how to use this technology at scale, the timing of that shift matters enormously.

Amara’s Law is catching up

Roy Amara observed that we tend to overestimate the impact of a technology in the short run and underestimate it in the long run. 2024 and 2025 were the short run. Everyone promised transformation; most delivered modest productivity. The backlash was predictable. But we are now entering the part of the curve Amara warned us about. The one where the technology quietly becomes real, and the organisations that have prepared the right infrastructure begin to pull ahead.

The mistake is to confuse the initial disappointment with the long-term trajectory. The models are ready. The question is whether the organisation is.

The hidden problem in most research organisations

Most research and consulting firms I talk to are quietly paying a tax they haven’t named. Every time a colleague opens an AI tool, they rebuild context from scratch. They upload the same methodological rules, retype the same house standards, re-explain the same analytical frameworks. Templates get re-attached. Brand voice gets re-specified. Work that should have taken minutes instead begins with a long manual preamble, every single time. And that’s assuming that your colleagues have done their prompt training and know where to find those templates!

Multiply that across hundreds or thousands of colleagues and you have a structural drag on productivity. The AI works fine. The institutional memory around it is missing and a Copilot prompt library ain’t going to cut it.

The dual failure mode

Organisations confronting this gap tend to react in one of two ways, and both fail.

Some clamp down. They restrict AI usage, require approvals for anything non-standard, and treat governance as the primary concern. The AI-savvy colleagues quietly route around the restriction. Personal subscriptions. BYOD. Private instances. The work moves underground, invisible to governance.

Others open the floodgates. They hand out licences, encourage experimentation, and hope quality takes care of itself. What emerges is inconsistency at scale. Different teams applying different logic to the same analytical problem. Outputs that vary in methodology, quality, and voice. A gradual erosion of trust in the technology itself, precisely because the outputs don’t reliably meet the standard.

A phrase I’ve used often this year captures the challenge: speed is easy; speed with trust is what sets you apart.

Institutional skills: the thesis

The research and consulting organisations that will benefit from AI at scale are the ones that codify their most valuable ways of working into governed, reusable skills that any colleague can invoke without needing to be a prompt engineer, and without reinventing the wheel.

Think of a skill as a permanent digital twin of a proprietary methodology. Once codified, it doesn’t need to be re-explained. It’s available by default. The AI behaves the way the organisation behaves, because the organisation has taught it how. Invisibly.

This shifts the locus of competitive advantage. It moves from “who has the best model” to “who has codified the most of their institutional expertise into the knowledge layer.” That’s a race defined by organisational discipline, not technological arms. It is, in Amara’s terms, the long-run advantage that compounds.

It also solves the dual failure mode. Well-governed institutional skills give AI-savvy colleagues a sanctioned path to move fast, while keeping quality and consistency high. The catalogue becomes the alternative to both over-restriction and free-for-all.

And critically this is only worth doing now that the underlying technology is reliable enough to bear the weight of institutional processes. Two years ago, the consistency wasn’t there. In 2026, it is.

The role: AI Knowledge Architect

Building this layer is a discipline we haven’t historically had a name for. At Euromonitor International we’re creating a role to lead it: the AI Knowledge Architect, reporting directly to me.

This person will own our canonical skill catalogue and decide what gets approved, and what standard the whole organisation builds to. They’ll hold formal sign-off authority. And they’ll apply a strategic test to every skill we consider: it must encode a proprietary Euromonitor way of working, or draw from our proprietary data. Anything a generic agent can already do on its own doesn’t earn a place.

It is equally a role for a senior technical builder who governs as a consequence of deep expertise, and for a governance leader who can get hands-on when it counts.

If this is your shape

We’re open to candidates in four locations:

• London
• Cape Town
• Vilnius
• Bengaluru

One small ask for anyone who’s read this far and is planning to apply. Please include the phrase “Institutional Compound” somewhere in your cover letter or application. It won’t affect your candidacy either way, but it’ll tell me you engaged with the piece. A signal that’s worth something in the era of application automation.

If you’ve been quietly solving this problem in your own corner of the research and insights world or you’ve been waiting for a role where the governance and the building are treated as the same job — I’d love to hear from you.

And if you know the right person but it isn’t you, the greatest help you can offer is a share.