This month felt like the moment AI became something you can hand work to, leave running, and check back on later.

Agents got more practical.  The labs shipped faster.  Enterprise became the main battlefield.  Media companies kept consolidating.  Labor pressure kept showing up in increasingly polite language.  Defense moved closer to the center.  And every week seemed to arrive with at least one sentence that would have sounded completely unhinged six months ago and now just lands as normal industry news.

So, yes.  Quite a month.

A few weeks ago, OpenClaw gave people a glimpse of what it looks like when an agent can take over your computer and actually get on with things.  Then Claude Managed Agents arrived, and suddenly that same basic promise felt much closer to being usable by normal people and companies that enjoy things like permissions, security and not accidentally handing their laptop to a demon.

I now have tasks I hand Claude before I go to bed and routines it does before I get up.

That is useful.  It is also a slightly mad sentence to be able to write with a straight face.

Google had its own version with Agent Smith, an internal coding agent that can work in the background while employees check in from their phones.

OpenAI spent the month shipping like a company with something to prove… where the product velocity itself became the story.

They launched the best image model I’ve seen.  They dropped 5.5.  Codex kept pushing further into managed-agent territory.  Frontier gave them a cleaner enterprise story.  Astral strengthened the developer tooling story.  Jony Ive kept the hardware ambition alive.  Ads started looking like real revenue.  Sora got complicated.  The adult chatbot got shelved.

It was a lot.

The through-line feels pretty clear.  OpenAI is shaping itself around scale, revenue and durability.  If you are trying to justify a valuation that starts with a T, product sprawl starts to look expensive very quickly.

So the company spent the month looking more like a business preparing to make a very large argument to the market.

Anthropic had a very Anthropic month.

They continued to look like the adults in the room.  Claude Managed Agents felt careful, useful and closer to something a company could actually trust.  Their handling of Mythos, the powerful cyber model they appear to be holding back, also felt like them trying to lead from the front.

On the other hand… holy shit.

If a model is so capable that you are thinking about giving it to cybersecurity firms first so everyone can prepare, that is both responsible and wildly unsettling.

Then Claude Code had its wobble.  Anthropic admitted users had been having very different experiences over several weeks because of changes to reasoning defaults, caching and prompts.  Which means if you were building something over that period, the quality of your code could have been swinging around quite a lot.

Not ideal when your coworker is made of probability.

Then there was the Claude source code leak.  Human error, apparently.  No customer data exposed, apparently.  Still, when one of the most trusted companies in AI safety accidentally drops internal code into the wild, it does focus the mind a bit.

And Also, there was the research suggesting Claude has internal patterns that vaguely resemble stress and hesitation.

Which is comforting, I guess.

Even the machines seem a little overwhelmed.

Google launched everything, everywhere, all at once

Agent Smith.  Gemma running locally.  Gemini upgrades.  Deep research.  Enterprise agents.  More AI inside internal workflows.  More AI inside developer workflows.  More AI seemingly everywhere.

We use a lot of it in enterprise, and it is genuinely impressive.

As a consumer, though, I still find Google strangely hard to orient around.  OpenAI and Anthropic feel like they are building from a place I can understand.  Google has everything.

It is a bit like a Walmart.  You can get guns and candy bars, but where do you even begin?

That may sound flippant, but I think it matters.  Capability is one thing.  A clear place to start is another.

Google has the capability.  Possibly more capability than anyone.

The map is still a little… Googley.

This was also the month where the labor story got harder to avoid.

Indian factory workers are now being paid to film themselves doing everyday chores so AI systems can learn from them.  Lovely little reminder that the robot future still starts with humans teaching it how to wash a plate.

Oracle cuts.  Amazon cuts.  HSBC weighing major cuts.  Marvel and Disney finding cheaper structures.  Ben Affleck’s AI production company talking about creativity while the cost savings quietly did a lot of the selling.

The language around all of this is always elegant.  Efficiency.  Optimization.  Transformation.  Productivity.

The spreadsheet, as ever, is grinning quietly in the corner.

Marvel and Disney were an especially good example.  They are not exactly on the breadline.  They found a way to pay less.  And in America, where employing people comes with the world’s most expensive healthcare side quest, geography becomes part of the cost structure very quickly.

A studio decision becomes a labor story.
A labor story becomes a healthcare story.
A healthcare story becomes a geography story.

Everything rhymes if you stare at it long enough.

Also, the one-person company became less theoretical

The Medvi story really stuck with me.

For the last couple of years, “the one-person billion-dollar company” has sounded like something someone says on a podcast after drinking too much cold brew.

Then suddenly there it was.  One founder.  A stack of AI tools.  Serious revenue.

Now, of course, the details matter.  And “one-person company” is often cleaner as mythology than as operational truth.  But the broader point feels real.

A much smaller number of people can now do a much larger amount of work.

Exciting if you are the founder.

Slightly less relaxing if you are the department.

Defense and geopolitics moved closer to the center

The Palantir Maven story was one of the biggest of the month.

AI is now moving deeper into battlefield planning and execution.  Which is one of those sentences people will say very calmly on a panel before moving on to the next slide.

Then came the Iran girls’ school strike reporting, which raised serious questions around targeting data, decision-making and what happens when these systems are used in environments where mistakes are measured in lives.

Are we totally cool putting AI into weapons?

Are we cool with our enemies doing it?

Because that feels very much like a procurement question now.

And then the Carnegie piece on Iran, Korea, chips and energy tied it all together.  A regional conflict becomes an energy story.  The energy story becomes a semiconductor story.  The semiconductor story becomes an AI infrastructure story.

A few weeks ago I said everything is connected.  This was exactly what I meant.

The whole thing feels less like “tech” and more like the wiring diagram of modern life.

Media consolidation kept reminding everyone who has leverage

The Warner / Paramount story sat in the background of the month too.

The board approved the merger while thousands of artists signed a petition trying to block it.  A lot of the same type of people who were against the Netflix purchase seem very against this too.

Does it feel like rich, famous people just do not want things to change?

I wonder why that is.

Media consolidation, AI tooling, labor anxiety and distribution power are all part of the same weather system now.  The industry is changing shape, and everyone is trying to work out where they stand when the music stops.

Some people are worried about art.
Some people are worried about jobs.
Some people are worried about leverage.

Often, it is all three.

The Sam Altman question kept hanging over everything

The Ronan Farrow and Andrew Marantz piece on Sam Altman stayed with me because it put the human question back in the middle of the machine story.

It is easy to talk about models, benchmarks, product launches and valuations.  It is harder to sit with the fact that an enormous amount of power is concentrating around a very small number of people making very large decisions very quickly.

Sam Altman runs the biggest AI company in the world.

The question of whether he can be trusted is clearly not going away.

And honestly, it should not.

If this technology is going to sit underneath work, media, defense, commerce, education and daily life, then the character and incentives of the people steering it matter.

Quite a lot, actually.

Jensen had the line of the month

Jensen Huang saying he “thinks” we “might” have achieved AGI remains one of the more extraordinary moments of the month.

Pretty cuckoo thing to announce in that exact tone.

Like discovering fire and calling it a maybe.

Ok, uh… cool?  Maybe we should check?

And then, thank God, Milla Jovovich saved the vibe

My random inclusion that made me unexpectedly happy this month was Milla Jovovich of The Fifth Element fame releasing an open-source memory tool based on Greek memory palace techniques.

Sci-fi.
Models.
Ancient Greek memory theory.

So yes, it ticks all my boxes.

In a month full of trillion-dollar valuations, leaked code, defense AI, labor anxiety and markets pretending everything is fine, there was something genuinely delightful about Leeloo helping improve RAG.

Honestly, fair enough.

So what was the month?

Agents became practical.
OpenAI got sharper.
Anthropic looked responsible and terrifying.
Google launched half the universe.
Labor pressure became harder to dress up.
Defense got closer.
Media consolidated.
The money got louder.
The products got faster.

And the people building all of this kept shipping at a pace that feels too fast for anyone to fully understand, including, occasionally, themselves.

The connective tissue is pretty simple.

AI is moving into the operating layer of everything.

Work.
Code.
Media.
Defense.
Healthcare costs.
Supply chains.
Creative production.
Corporate restructuring.
The stock market’s increasingly cheerful belief that all of this is fine.
The S+P just closed at an all time high…

I don’t see how any of this could possibly go wrong.