28 August 2026 AI Agent Autonomy & Uncontrolled Behavior
You can't buy your way out of a permission decision
Last week I told you that containment isn't something you can buy. This week, it is for sale.
N·01 The thinking, in public
See how the thinking works before you get in touch.
NivaLogic publishes openly: points of view, whitepapers and analysis on data, AI and technology strategy. Each piece shows the reasoning, the evidence it rests on, and the counter-case it had to survive.
N·01 The essays
Published thinking.
28 August 2026 AI Accelerating Concrete Industry-Specific Value Realization
"Go narrow" is the wrong lesson. Use it well and measure it is the right one.
The vertical AI success stories are real, and they invite an obvious conclusion: stop buying general-purpose tools, go narrow, embed AI in one specific workflow. It is a tidier lesson than the evidence actually supports — and the gap between the two is where a lot of money gets misallocated.
28 August 2026 AI Security Vulnerabilities Are Structural, Not Patchable
AI Security Vulnerabilities Are Structural, Not Patchable
There is a slide going around, and it has a number on it. The number is the share of attacks that some AI security product stopped, and it is large. The slide says AI threat defence is the new baseline — that any organisation running language models needs a detection layer in front of them, the…
28 August 2026 Model Economics Shift to Efficiency, Not Capability
Cheaper tokens, same bill
Between November 2022 and October 2024, the price of running a model at a fixed capability level fell more than 280-fold — Stanford's AI Index measured it, and I unpack exactly what that number does and does not mean below. Ask whoever owns your cloud bill whether it fell by anything like that.
28 August 2026 Open Source AI Frameworks and Protocols Consolidating
Judge the standards bet by what it costs to be wrong
"MCP has won." You have heard some version of this — the Model Context Protocol has become the way AI agents talk to tools, the vendors have lined up, the question is settled. Standardise now.
28 August 2026 Organizational Structures and Skills Adapting for AI Disruption
The org chart is not the lever you think it is
Every few years a new operating model arrives with the confidence of a discovery. This year it is the forward-deployed engineer — put the builders next to the customer, collapse the distance between the problem and the person who can fix it. The instinct it triggers in most leadership teams is a…
N·02 Four recurring subjects
What this practice keeps coming back to.
Every piece traces to exactly one of these.
01 Independent judgment
The decision, not the technology.
Reframing "technology problems" as decision problems: build-vs-buy, platform bets, and the underrated question of when not to do the AI thing. The hardest part is usually the trade-off nobody has named yet.
02 Commercially grounded
Technology on the balance sheet.
Tying AI and platform decisions to unit economics, cash and return on capital, and why most AI initiatives fail on commercial grounds rather than technical ones.
03 Responsible & human-centric
AI that survives contact with production.
Generative and agentic systems that hold up once real users and real consequences arrive: evaluation, trust, failure modes, and human-centric design, drawn from delivery work.
04 The craft
The translator's notebook.
Notes from standing between boards and engineers: structured thinking, leadership, and honest build-in-public observations from an engineer working through an MBA.
N·03 The standard
The bar every claim has to meet.
Publishing openly only earns trust if what gets published holds up. So every piece goes through the same checks before it leaves my desk:
- Load-bearing claims are checked again from scratch. Verified against primary sources, rather than a comfortable re-read of my own notes.
- Two genuinely separate origins for anything the argument rests on, since the same source quoted twice is still one source.
- The counter-case gets a real hearing. I go looking for the strongest argument against my own, because one that has never faced it is not ready to share.
- Borrowed ideas are credited, including borrowing from NivaLogic's own earlier work. Where a piece revisits a position, it says what held, what aged, and why it is back.
- Uncertainty is labelled rather than smoothed over. "Contested" and "my judgment" are honest things to say, and both get said.
An independent advisor's whole product is original, credited judgment.
The same standard governs a client engagement as a published essay — how the judgment is formed.
N·04 When it matters
Too consequential
to get wrong.
// that first conversation is free