Let a Thousand Agents Bloom

How we build will change what we build. What happens when our collaborators number in the thousands? Also: Sam Altman invests in brain-computers, Github CEO exits, Apple moves into

We’re entering the age of cognitive abundance… if your brain can handle it.

For the past two years, it’s felt like drinking from the firehose. Some weeks, it feels like a stress test for neuro-plasticity. The constant learning and unlearning of mental models and new technical approaches. The rewiring of thought patterns. That vertigo when you realize: my brain didn’t think this way a year ago.

The good/bad news: I don’t think it’s going to slow down. This moment definitely feels like a pre-Internet/post-Internet rewiring of our brains into new behavior… like how we now search instead of remember, or reference instead of recite.

One of the biggest rewiring I see on the horizon: we will shift from making decisions to designing the systems that make decisions.

More systems thinker. More organizational design. Even when you’re a human team of one.

The rise of the orchestra

What used to take a small team can now be done by one person, if that person (or their AI) knows how to conduct an orchestra of AI agents.

Early research into orchestration explores strategies like root agents that dynamically spawn other agents, and meta-agents that coordinate entire swarms.

You can see this in Anthropic’s Claude Research, Google’s Gemini Deep Research, and tools like Manus Wide Research. All point to the same shift: multi-agent architectures where many agents run in parallel. Instead of a single model grinding through tasks sequentially, these ensembles coordinate, check each other’s work, and refine outputs in real time.

Parallel swarms aren’t a silver bullet. Research tasks, or what Anthropic tags as “breadth-first tasks”, parallelize beautifully. Their research found that coding tasks often don’t, since they are more sequential. Too much depends on the previous step to run well in parallel.

Where parallel agents shines:

• Complex coordination: Many moving parts that benefit from concurrency.

• Perspective diversity: Multiple “minds” exploring the same problem from different angles.

• Self-improvement: Agents that critique each other’s work along the way

How we build will define what we build. Once orchestration becomes common, the world will look very different.

What happens when projects that once took months will compress into days? When decision-making routinely blends multiple expert perspectives in real time?

And the systems themselves will start designing new systems?

The real change isn’t just speed or scale; it’s that our default working mode will shift from doing things ourselves to shaping the systems that do them for us.

Enjoy the weekend,

Tara

In next week’s Product Patterns, I’ll break down the emerging playbook for Multi-Agents: the core product patterns, the business implications, a first cut at a market map, and where I think capital will flow next.

The market map and deep-dive analysis will be available to Operator subscribers, my paid tier for founders, operators, and investors who want to stay ahead of these shifts before they hit the mainstream. It’s free to explore until launch day next month.

My nearly-5 year old can’t quite read - but he can vibe code. In ChatGPT’s conversation mode, we wrote a short spec for a “piggy bank” web app.

Minutes later, using Cursor and GPT-5, we had the first MVP live.

1. OpenAI brings GPT-4o back as a default for all paying ChatGPT users, Altman promises ‘plenty of notice’ if it leaves again: OpenAI is once again making GPT-4o a default option for all paying users. Paying ChatGPT subscribers will also get a new “Show additional models” setting on by default that restores access to GPT-4.1, o3 and o4-mini, the latter two reasoning-focused LLMs. Altman’s latest update adds new controls to the ChatGPT interface: users can now choose between “Auto,” “Fast,” and “Thinking” modes for GPT-5.

2. Google’s Gemini AI will get more personalized by remembering details automatically: Google is rolling out an update for Gemini that will allow the AI chatbot to “remember” your past conversations without prompting. Users can also preserve privacy through a new “temporary chats” option in Gemini.

3. Perplexity offers to buy Google Chrome for $34.5 billion: Perplexity has offered to buy Google Chrome for $34.5 billion – a bid that’s far more than the AI search startup itself is valued at, according to reports from The Wall Street Journal and Bloomberg. The startup sent the unsolicited bid just months after Perplexity said it would buy Chrome if the government forces Google to sell its browser. Google hasn’t indicated that it would sell Chrome at any price, and so far, the court hasn’t ordered a sale.

4. Sam Altman, OpenAI will reportedly back a startup that takes on Musk’s Neuralink: Sam Altman is in the process of co-founding a new brain-to-computer interface startup called Merge Labs and raising funds for it with the capital possibly coming largely from OpenAI’s ventures team, unnamed sources told the Financial Times. The startup is expected to be valued at $850 million. Merge Labs will compete with Elon Musk’s Neuralink, which is developing computer interface chips designed to be implanted in the brain

5. Claude can now process entire software projects in single request, Anthropic says: Anthropic announced that its Claude Sonnet 4 artificial intelligence model can now process up to 1 million tokens of context in a single request – a fivefold increase that allows developers to analyze entire software projects or dozens of research papers without breaking them into smaller chunks. With the new capacity, developers can load codebases containing more than 75,000 lines of code, enabling Claude to understand complete project architecture and suggest improvements across entire systems rather than individual files.

6. Microsoft absorbs GitHub into CoreAI division as CEO plans exit: Microsoft is poised to fundamentally reshape GitHub following CEO Thomas Dohmke's resignation. This transition ends GitHub's semi-independent status and integrates the platform into Microsoft's newly formed CoreAI division, focused on artificial intelligence development.
   

7. Apple plots expansion into AI robots, home security and smart displays: Apple is plotting its artificial intelligence comeback with an ambitious slate of new devices, including robots, a lifelike version of Siri, a smart speaker with a display and home-security cameras. A tabletop robot that serves as a virtual companion, targeted for 2027, is the centerpiece of the AI strategy, according to people with knowledge of the matter.

1. The road to artificial general intelligence (MIT)

2. AI Agents make up a third of all search traffic toward brands, report says (CNET)

3. Physicists create quantum radar that could image buried objects (MIT)

4. Sutter Hill’s Mike Speiser does a rare interview about the firm (Goldman Sachs)