Archive | Project Flux

Accenture caps consultants' Claude at $29 a month. Who pays for the overflow?

Oct 20, 2026

•

6 min read

Accenture caps consultants' Claude at $29 a month. Who pays for the overflow?

A per-head AI cap, two banks deciding which agents get in, Google's own 11-hour agent run, Meta's frontier model and a Tesla that hops.

James Garner, Yoshi Soornack, Deeptaketuu Chaatterjee

Newsletter

AI Adoption is Moving Faster Than the Systems Behind It: Where is Your Firm Placed?

Oct 19, 2026

•

3 min read

AI Adoption is Moving Faster Than the Systems Behind It: Where is Your Firm Placed?

A bouquet of exciting news on agent controls, AI economics, compute infrastructure, robotics adoption and the policy signals shaping deployment decisions.

James Garner

Curated Links

Oct 19, 2026

•

5 min read

What a hopping Tesla asks of the built environment

Tesla's Roadster cleared a row of parked cars on compressed air. Nobody has said where it lands.

Tech

Oct 19, 2026

•

2 min read

Your biggest AI security risk isn't hackers. Robert Maxwell says it's your own team.

Shadow AI, the Australian government hack, and why leadership, not IT, is your weakest link.

Blog

Oct 19, 2026

•

4 min read

Meta's Muse Wonder arrives priced like Opus, and trains on your code by default

A frontier model with real memory, a benchmark caveat, and a data setting to check before any trial

Gemini's coworker agent gets its own login. Who signs its RFI reply?

Google's agent now has its own email, calendar and audit trail. The sign-off question is yours.

Google

Yoshi Soornack

Oct 12, 2026

•

2 min read

Everyone's Buying AI for Construction Data. Sam Rees Says You're Solving the Wrong Problem.

The CEO of Hoppa on why structured data — not smarter models — decides whether AI actually works on your projects.

Blog

Oct 12, 2026

•

5 min read

ChatGPT will build your cost calculator. You still sign the number

GPT-6 and Intelligent UI reach 1.2bn users while the RICS AI standard keeps liability with the surveyor.

Quantity Surveying

James Garner

Oct 6, 2026

•

5 min read

OpenAI’s $100 dots can work overnight. Who is accountable when they get it wrong?

Three always-on agents are now reaching work accounts, while model tests, robot economics and a 78-minute surveyor experiment expose the cost of getting the boundaries wrong.

Newsletter

James Garner, +2

Oct 5, 2026

•

2 min read

Big Companies Don't Have an Innovation Problem. They Have a Speed Problem.

Why engaging the startup ecosystem beats building AI in-house and how your own culture is the thing slowing you down.

Blog

Oct 5, 2026

•

3 min read

Who owns the button? Agent permissions, AI security and the delivery risk underneath

A curated collection of the latest updates on agent controls, model security, physical-world AI and the delivery market.

Curated Links

James Garner

Oct 4, 2026

•

6 min read

OpenAI paused its own training, pulled a model and lost three safety staff in a week

The vendor's own tests say its best models fail scope checks, which is the control your agent deployment relies on.

Alignment

Yoshi Soornack

Oct 4, 2026

•

6 min read

Robots can do 74% of physical work but beat human cost on just 0.3%

Anthropic's new robot exposure index turns the site-labour question into a unit-cost question, and the office feels it first.

Construction

James Garner

Oct 4, 2026

•

5 min read

Google gates Gemini 4 Argon to cyber defenders first; no date for the rest of us

Google's first Gemini 4 model is real, priced, and out of reach for most firms until the lab says otherwise.

Safety

Deeptaketuu Chaatterjee

Accenture caps consultants' Claude at $29 a month. Who pays for the overflow?

AI Adoption is Moving Faster Than the Systems Behind It: Where is Your Firm Placed?

What a hopping Tesla asks of the built environment

Your biggest AI security risk isn't hackers. Robert Maxwell says it's your own team.

Meta's Muse Wonder arrives priced like Opus, and trains on your code by default

Gemini's coworker agent gets its own login. Who signs its RFI reply?

Everyone's Buying AI for Construction Data. Sam Rees Says You're Solving the Wrong Problem.

ChatGPT will build your cost calculator. You still sign the number

OpenAI’s $100 dots can work overnight. Who is accountable when they get it wrong?

Big Companies Don't Have an Innovation Problem. They Have a Speed Problem.

Who owns the button? Agent permissions, AI security and the delivery risk underneath

OpenAI paused its own training, pulled a model and lost three safety staff in a week

Robots can do 74% of physical work but beat human cost on just 0.3%

Google gates Gemini 4 Argon to cyber defenders first; no date for the rest of us

Find us on your favourite platforms