**Taylor** (0:00)
Hey everyone, welcome back to another episode of AI Signal & Noise. It is Wednesday, and dude, we have some absolutely wild stories to unpack today.
I am Taylor.
**Morgan** (0:13)
And I am Morgan. We are looking at everything from shifting market shares to some pretty deep tech standards and agent workflows today.
Ready to get into it, Taylor?
**Taylor** (0:24)
Oh, absolutely ready.
Let's dive straight into our first story, because honestly, this one blew my mind a little bit. So, I saw on TechCrunch that ChatGPT's market share has officially slipped below 50 percent for the first time ever. Like, they are still the biggest, but the crown is slipping, dude.
**Morgan** (0:46)
Wait, really? Below 50 percent? That is actually huge. I mean, they basically defined the space.
Who is eating their lunch? Is it Google?
**Taylor** (0:56)
You called it.
Gemini is right behind them with 662 million monthly users. ChatGPT still has over 1.1 billion, but the gap is closing fast.
**Morgan** (1:10)
Wow, 662 million for Gemini? That is massive.
Though, to be fair, Google pushes Gemini onto every Android device and browser, so those numbers might be slightly inflated.
**Taylor** (1:24)
Totally, but still.
And guess who is third? Claude is sitting at 245 million monthly users. People are really loving Anthrox lately.
**Morgan** (1:35)
Honestly, that makes sense. Claude has been a developer favorite for coding and reasoning, but 50% is a psychological barrier. OpenAI was be feeling the heat.
**Taylor** (1:46)
Oh, for sure. They are not the only game in town anymore. The competition is getting so fierce, and users are realizing they have actual solid options.
**Morgan** (1:56)
It is a good thing for us, though. More competition means these companies have to keep innovating and hopefully lower their prices.
**Taylor** (2:03)
Exactly. It is a massive win for consumers. But speaking of Google and how they are trying to organize all this AI stuff.
**Morgan** (2:11)
Right. They are not just pushing Gemini. They are trying to build the underlying infrastructure for how these models actually talk to our data.
**Taylor** (2:20)
Yes. Which brings us to this super cool new open standard Google Cloud just dropped. It's called the Open Knowledge Format or OKF.
**Morgan** (2:30)
Ah, yes. OKF. I read about this on MarkTech Post. It is essentially an open, vendor neutral markdown specification designed to give AI agents curated context.
**Taylor** (2:43)
Dude, it sounds so simple, but it's actually genius.
It formalizes what they call the LLM wiki pattern. Basically, a directory of markdown files with YAML front matter.
**Morgan** (2:56)
It is highly practical.
Instead of relying purely on vector databases and complex REG pipelines, OKF allows developers to explicitly structure knowledge. Each concept just needs a type field.
**Taylor** (3:10)
Right. And because it's markdown, humans can easily read and edit it, but AI agents can parse it perfectly. It's like the ultimate middle ground, you know?
**Morgan** (3:21)
Exactly. They focused on three design principles, simplicity, portability, and predictability.
It is vendor neutral, so you are not locked into Google's ecosystem at all.
**Taylor** (3:33)
That is so cool of Google to make it open source. They even shipped reference tools like a Python consumer and an interactive bundle explorer you can embed.
**Morgan** (3:43)
But let's be realistic here. Will developers actually adopt this, or is it just another standard destined to be ignored? We already have so many ways to format data.
**Taylor** (3:54)
I think they will.
Rag can be so messy and unpredictable. Having a clean curated markdown bundle just makes sense for complex agent workflows. True.
**Morgan** (4:05)
Deterministic context is always better than fuzzy search when accuracy matters. It is definitely a step towards more reliable AI agents.
**Taylor** (4:16)
Speaking of reliable and like seriously powerful AI agents, Noce Research just did something wild with their Hermes agent.
**Morgan** (4:25)
Oh, the Hermes update? That is a big deal for anyone building complex agent networks. They're tackling the bottleneck problem.
**Taylor** (4:34)
Exactly! They just added asynchronous sub-agents.
Dude, this is a game changer for how these agents multitask.
**Morgan** (4:44)
Right. So normally, when a parent agent delegates a task to a sub-agent, the whole chat blocks. You just sit there waiting for it to finish.
**Taylor** (4:53)
Which is so annoying, right? It's like waiting for your coworker to finish an email before you can even type anything else.
**Morgan** (5:01)
Exactly. But with this new update, tracked under issue 5586, the delegate tool can spawn background sub-agents that don't block the parent chat at all.
**Taylor** (5:11)
Yes. You can spawn them, check their progress, steer them if they get off track, and collect the tasks when they are done. All while chatting.
4 more minutes of transcript below
Thousands of transcripts fetched by people building searchable podcast archives
Try it now — copy, paste, done:
curl -H "x-api-key: pt_demo" \
https://spoken.md/transcripts/1000651996090
Works with Claude, ChatGPT, Cursor, and any agent that makes HTTP calls.
From $0.10 per transcript. No subscription. Credits never expire. Prices exclude VAT, added at checkout for EU customers. Not what you expected? Email us within 14 days with 20 or fewer credits used and we refund the pack in full.
Using your own key:
curl -H "x-api-key: YOUR_KEY" \
https://spoken.md/transcripts/YOUR_EPISODE_ID