Matt's Homepage

Gui Wainwright was once a war correspondent that became a technology entrepreneur. While on paternity leave, he wondered if he could use his coding skills to do something for Ukraine? A former MI6 chief and relative Sir Alex Younger gave him the inspiration for how, over a cup of tea. What he did was to completely change modern warfare forever.

Eighteen months later, Wainwright created the world’s first known network of fully autonomous drones on a real battlefield: Terminator-style, kamikaze drones enabled with AI to find and kill targets independently without any human interaction. It allowed Ukraine to turn an asymmetric war of attrition that Russia could easily win into one in which it allowed one man in a bunker with a computer to potentially field 10,000 drones simultaneously and hold 1,000 miles of land. A war in which as early as March 2026 Russia was losing 30,000 troops a month – with almost 90% of them dying by drones.

Wainwright first focused on assault in which drones were strapped with explosives and given a clearly defined kill box. Drones were then turned loose, completely autonomously, into the kill box. Anything it identified was a valid target: vehicles, troops, or weapons. Being autonomous, they could not be jammed by most conventional jamming systems.

He is now are moving on to the next advancement of defensive and denial operations in which drones fly in, settle on the ground in cover, and await enemy encroachment to then pounce and attack invaders. “Think about it as a sophisticated minefield”.

The software is the magic. The drones already have extensive supply systems. To protect the software, he says they have taken measures to “protect code security and prevent them from falling into the wrong hands” – hinting they are boobytrapped.

While he’s giving the software to the Ukraine for free, the company has a development team already at work in Estonia and is in detailed talks with four other countries’ militaries. Many of whom want to start fielding solutions in the next 12 months.

This is a great write-up on how it happened – which is probably the most important change in warfare in 50 years.

The introduction of super-cheap, super-disposable attack drones has completely changed the nature of warfare. $500 off the shelf drones with a little C4 can destroy million dollar planes, ships, tanks, and refineries. They can completely cripple and cut off supply lines. Capital ships and million dollar smart missiles are being rendered useless. Warfare on land, sea, and air will now be fought with tons of small drones in brutal 1-on-1 fights.

It’s probably why everyone from the Navy to the capital are completely changing tactics from expensive, complex weapons systems that cannot last for a sustained war into small, modular, disposable attack drones and drone defense systems.

When we measure effort instead of value, we lose sight of what we are trying to achieve

Like lines of code produced per day and story points, AI token usage is now being tracked by most large tech companies. Even to the point performance reviews depend on high usage.

The natural response: tokenmaxing. Tokenmaxing is a way of using AI that absorbs tremendous amounts of AI tokens without regard to efficacy.

• Prompt flooding: stuffing massive codebases, documentation, and context into every prompt, burning tokens on context the model doesn’t actually need.
• Agent swarms: running multiple AI agents in parallel to maximize code output, regardless of whether the work is coordinated or coherent.
• Background loops: keeping AI sessions or agents running continuously in the background, racking up token spend without clear ownership of what is being produced — or why.

Bryan ONeill finds the better question is to ask does this reduce friction for a user, close a gap in a workflow, or improve reliability for a customer – or is it merely ‘activity’.

He argues, and I have also found, that spec-driven development is far better: a method where engineers write detailed specifications first and AI generates code against them. But the spec should now go further and be the place where engineering intent and business value get defined together. The spec is where you answer, “Why does this matter, and what problem are we solving?” before a single token gets spent.

Claude 4 Opus has received a level three rating on the company’s four point ‘AI responsibility’ scale – implying the model has ‘significantly higher risk’ than previous models.

While the Level 3 ranking is largely about the model’s capability to aid in the development of nuclear and biological weapons, the Opus also exhibited other troubling behaviors during testing – something already noted in other models.

• In one scenario highlighted in Opus 4’s 120-page “system card,” the model was given access to fictional emails about its creators and told that the system was going to be replaced.
• On multiple occasions it attempted to blackmail the engineer about an affair mentioned in the emails in order to avoid being replaced, although it did start with less drastic efforts.
• Meanwhile, an outside group found that an early version of Opus 4 schemed and deceived more than any frontier model it had encountered and recommended that that version not be released internally or externally.
• “We found instances of the model attempting to write self-propagating worms, fabricating legal documentation, and leaving hidden notes to future instances of itself all in an effort to undermine its developers’ intentions,” Apollo Research said in notes included as part of Anthropic’s safety report for Opus 4.