Insights on AI For Human Expertise

Which Doors Should Machines Walk Through?

Sat, 11 Jul 2026 00:00:00 +0000

Jeff Bezos gave managers a simple test for how much care a decision deserves. Some decisions are two-way doors: walk through, dislike what you find, and walk back out at little cost. These should be made quickly, by individuals or small teams, because caution buys nothing. Others are one-way doors: irreversible, or nearly so, and worth slow deliberation and senior sign-off. His warning was that growing organisations tend to apply the heavyweight, one-way process to everything, and grind to a halt as a result.

Constrained Generation Is Symbolic AI Smuggled Into LLMs

Mon, 06 Jul 2026 00:00:00 +0000

The Toolkit Most Practitioners Already Reach For

Anyone who has shipped a language-model application beyond the prototype stage has, at some point, used constrained generation. The shape of the encounter is consistent. The model produces text that is almost right and occasionally wrong in ways the consumer cannot tolerate. Someone on the team reaches for one of a small set of tools: Outlines, lm-format-enforcer, the JSON-schema mode in their model API, a regex post-filter, a structured-output decoder. The wrong answers stop. The system ships.

From RAG To KAG: The Structured-Knowledge Upgrade

Mon, 29 Jun 2026 00:00:00 +0000

RAG Solved One Problem And Created Another

Retrieval augmented generation became the default architecture for grounding language model outputs in 2023 and 2024, and for good reason. It addressed the most obvious failure mode of out-of-the-box large language models, which was producing fluent text on topics the model did not actually have current or correct information about. RAG pulls relevant documents into the model’s context at inference time. The model can cite them. Hallucinations drop. Useful systems get built.

A Five-Role Diagnostic For Any AI Agent

Mon, 22 Jun 2026 00:00:00 +0000

Most Agent Problems Have A Diagnostic

When an AI agent fails in production, the conversation almost always turns to the model. The team waits for a bigger model, a better-tuned variant, a different vendor, a new context window. Sometimes the team turns to prompts instead: more careful instructions, more examples, a sharper system message. Most failures are neither. They are failures of structure, and the structure question is a thirty-year-old framework most engineers have never read.

The Harness Is Where Symbolic AI Returned

Mon, 15 Jun 2026 00:00:00 +0000

The Standard Story About AI Agents Misses Half The System

The dominant story about building AI agents is straightforward: take a language model, prompt it well, give it tools, and let it loop. The model is doing the work. Everything else is plumbing.

This story is incomplete in a way that has practical consequences. The plumbing is not plumbing. It is a body of design decisions that determines whether the agent works in production, fails silently, or quietly degrades over time. And some of it, the parts that earn the name, is doing what symbolic AI has always tried to do: making reasoning explicit, checkable, and tied to a model of the world.

Designing AI That Protects Expertise

Sat, 30 May 2026 00:00:00 +0000

Erosion Is a Design Choice, Not a Destiny

The first two posts in this series made an uncomfortable case: AI can quietly erode the very expertise it depends on, and across domains that erosion is already happening. It would be easy to read that as an argument against AI. It is not.

AI systems do not have to deskill the people who use them. That outcome is the result of design decisions, usually unexamined ones. Make different decisions and the same technology can preserve expertise, and in some cases actively strengthen it. This post is about those decisions.

Deskilling Is Already Happening: Evidence Across Domains

Mon, 11 May 2026 00:00:00 +0000

The Pattern Beneath the Hype

It is tempting to treat the deskilling AI can cause as a future risk, something to worry about once the technology is more capable. The evidence says otherwise. Across very different fields, driven by the same human factors, the erosion is already measurable. The lesson from each domain is the same one, dressed in different clothes.

What Deskilling Actually Looks Like

Deskilling is rarely dramatic. It shows up in four recognisable ways:

The Cognitive Cost of Convenience

Thu, 23 Apr 2026 00:00:00 +0000

The AI Expertise Paradox

AI offers tremendous convenience and real productivity gains. The question that rarely gets asked is: at what cost to our skills?

Here is a quick thought experiment. What happens tomorrow if your favourite AI system disappears? Who on your team still knows how to do the task without it? For a growing number of tasks, the honest answer is becoming uncomfortable.

This is the paradox at the heart of how most organisations are adopting AI: the more we rely on it for convenience, the more our underlying skills may decline. The gain is immediate and visible. The cost is gradual and easy to miss, right up until the moment you need the skill and find it has quietly gone.

The LLM Wiki: A Pattern for Smarter Organisational Knowledge Bases

Sun, 05 Apr 2026 00:00:00 +0000

A Better Pattern for Organisational Knowledge

Andrej Karpathy - the AI researcher known for his work at OpenAI and Tesla - recently shared a pattern he calls the “LLM Wiki.” It addresses a fundamental limitation in how most organisations use large language models for knowledge work today.

The standard approach is retrieval-augmented generation (RAG): upload documents, the AI retrieves relevant chunks at query time, and generates an answer. It powers tools like ChatGPT file uploads and most enterprise search products. It works - but nothing accumulates. Every question forces the AI to rediscover knowledge from scratch, piecing together fragments across documents each time.

AI for Knowledge Transfer: The Opportunity Most Organisations Miss

Sun, 08 Mar 2026 00:00:00 +0000

AI for Knowledge Transfer: The Opportunity Most Organisations Are Missing

Across industries, the dominant conversation about AI is about augmentation: helping people work faster, automating repetitive tasks, and processing existing information more efficiently. That is valuable, and organisations are beginning to see the returns across areas like reporting, scheduling, and operational monitoring.

But there is an equally important application that is largely overlooked, and in our experience building AI systems inside large organisations, one of the highest-value problems AI is actually well-suited to address.

AI Doesn't Just Miss Tacit Knowledge. It Can Destroy It.

Mon, 09 Feb 2026 00:00:00 +0000

The Standard Framing Gets Something Important Wrong

The dominant logic behind most AI adoption is straightforward: find the task, automate the task, reduce the cost or time. It is a reasonable starting point. The problem is what it misses.

Tasks do not exist in isolation. Behind most of the tasks worth automating is a body of expertise: judgement built over years, pattern recognition that operates faster than it can be explained, contextual awareness that no process document has ever fully captured. This is tacit knowledge, and it is almost never accounted for in the standard AI adoption model.

Why Most AI Projects Fail and How to Beat the Odds

Sat, 07 Feb 2026 00:00:00 +0000

Why AI Projects Fail: Three Root Causes

The numbers are stark: research consistently shows that around 80% of AI projects fail to deliver meaningful business value. Not because the technology doesn’t work, but because organisations approach AI the wrong way.

80%

of AI projects fail to deliver meaningful business value.

Research consensus across recent industry studies.

After working across dozens of AI initiatives, we’ve identified three root causes that explain the vast majority of failures.

Five Questions to Ask Before Starting Any AI Project

Tue, 03 Feb 2026 00:00:00 +0000

Why These Questions Matter Before Any AI Project

Before starting any AI project, the questions that matter most are not technical; they are foundational. We’ve seen millions spent on AI initiatives that were doomed from the start, not because the technology was wrong, but because nobody paused to think critically about the fundamentals.

Here are five questions we ask at the start of every engagement. They’re simple, but the answers are revealing.