Insights
Evidence, arguments and field notes on augmented work.
We publish what we actually find — including the results that argue against using AI.
Featured · Evidence
The jagged frontier: why AI helps on one task and hurts on the next
In a controlled study of 758 consultants, AI raised quality by around 40% on tasks inside its capability frontier — and made people 19 percentage points more likely to be wrong on a task just outside it. Here is how to tell the difference before you commit.
Dileep · 12 March 2026 · 8 min read
Point of view · 6 min
From vibe to production grade: what the last 10% actually costs
A working prototype and a reliable operation are separated by everything that matters: evaluation, failure modes, integration and ownership.
Field note · 10 min
Running multi-agent systems where money is at stake
Verification, guardrails and the cost of being wrong at machine speed, from a system that bids in a live market.
Foundations · 9 min
Engelbart was right: the tool is the smallest part
What a 1962 report predicts, with uncomfortable accuracy, about your 2026 AI programme.
For executives · 5 min
Six questions to ask before approving an AI business case
Including the one almost no proposal answers: what does checking the output cost?
Evidence · 7 min
When human + AI is worse than either alone
A meta-analysis of 106 experiments, and what it says about designing the handoff between them.
Field note · 8 min
Why the integration layer decides the outcome
Notes from replacing legacy B2B middleware while the business that depends on it keeps running.
One argued piece a month. No newsletter filler.
New research on augmented work, read and interpreted for people who have to decide something.
[Connect this form to your mailing list provider before launch.]