Adam Thyer

Founder at Pioneering a systems thinking approach to agentic development, education, and life.

[Adam Thyer @AdamThyer]

@theo @thomasboomcom @typelessdotcom @WisprFlow WhisprFlow is terrible! But then again, so is every other speech to text app I've used. Like you @theo, I just want something that accurately transcribes what I say. If it needs polishing, I can do that myself, or get a smart model to do it.

[Adam Thyer @AdamThyer]

@thomasboomcom @theo @typelessdotcom @WisprFlow $30 month. Ouch!

[Adam Thyer @AdamThyer]

@CompleteSkeptic @notkevinzhang

[Adam Thyer @AdamThyer]

@CompleteSkeptic I put myself on the waitlist 16 Sept. I received an email saying, "You're in!" 17 Sept. When logging in today, I'm told "Whoops, we're full". I understand not letting new people off the waitlist. But rejecting someone who was accepted several days ago is ludicrous!

[Adam Thyer @AdamThyer]

@aquavoice @WisprFlow I just tried it briefly and it slowed my Mac Mini to a crawl. So not for me. I really wish someone could build a decent voice transcription service.

[Adam Thyer @AdamThyer]

@WisprFlow was good for the first couple of weeks using it. Then they swapped the good model for an incompetent one. Today, they seem to have downgraded even further. Anytime I've contacted support, they ignore me. What's better?

[Adam Thyer @AdamThyer]

@aquavoice @WisprFlow You almost had me. But a thousand free words is not sufficient to test it. You request screen recording permissions without a detailed disclosure of what happens to to the data. You offer a discount with a .edu email, but no obvious means to register by email.

[Adam Thyer @AdamThyer]

@prasenx Anyone can vibe code. But nobody should vibe code (software engineers included) if they need production grade code.

[Adam Thyer @AdamThyer]

@NoemiTitarenco Humans are also next token predictors.

[Adam Thyer @AdamThyer]

@julien_c Diogo says "no" because they have upcoming capabilities that go beyond that definition.

[Adam Thyer @AdamThyer]

@mignano I think about user facing agents as coworkers. Sometimes, you interact face to face (via a bespoke GUI) and sometimes via 3rd party communication platforms.

[Adam Thyer @AdamThyer]

@AndrewYNg This is a complex topic. I think safety has to be the #1 priority. However, the response needs to be calm and rational. While I fear the potential dangers, I also fear paranoid pandering ignorant regulation. Thanks for this contribution to the debate 🙏

[langfuse.com @langfuse]

@marcklingen sat down to demo all langfuse features in a single video.

Agent observability, monitoring, evals, prompt management, and what is special about Langfuse.

Great place to start if you are new, or catch up on latest changes.

https://t.co/3lSorqdNme

[Adam Thyer @AdamThyer]

Yup. I love it when entrepreneurs create new product categories ♥️

[Nicolas Bustamante @nicbstme]

Jev is a textbook example of Clayton Christensen’s Innovator’s Dilemma playing out in AI.

It is commoditizing the bottom of the ML market: classifiers, routers, scoring and context triage. For frontier LLM labs, these workloads are not even worth pursuing. Revenue per decision is tiny, margins are razor-thin, and their entire infrastructure is optimized around selling expensive reasoning and token generation.

They cannot simply put a smaller LLM on fast inference chips like Cerebras and compete. Jev uses fundamentally different infrastructure, built to return typed probabilistic decisions so it’s a pure disruption

[James Ward @JamesWard]

The reason why developers, myself included, are excited about Jev is because it is a programmable AI primitive.
GenAI / LLM inference is only programmable in a very awkward, indirect way - mostly it is just smart autocomplete.

[Adam Thyer @AdamThyer]

I've read too many stupid tweets about Jev, this is one of the smart ones 👏

[Justin Schroeder @jpschroeder]

I’ve put a ton of tokens through Jev now, and...I have thoughts.

1. The big labs lost the script

TypeSafe’s intro video says, “We’re building prod, not God” — and can I just say? Hell yeah. @typesafeai built something for engineers like me, so I can build products for people like your mom (she says hi btw). They’re not trying to create “Machines of Loving Grace”; they’re trying to build things that let OTHER HUMAN BEINGS create new types of products.

The proof is in the pudding. When ChatGPT shipped, we were all blown away by…ChatGPT. When Jev shipped, we were all blown away by what everyone was making with it.

I cannot tell you how refreshing this is. It takes a sincere form of humility to believe that you, the creator of a technology, will not be the best at deploying the technology into the marketplace. The big labs, especially Anthropic, have proven not to have this humility…and now everyone hates AI. Thanks, Dario.

I guess what I’m saying is: Diogo for president.

2. Ultra-smart classifiers were genuinely a missing primitive.

I don’t know how this got missed. Diogo calls it “machine-native intelligence,” and that’s really what it is. I cannot tell you how much bending and twisting I've had to do with LLMs to get them to act like a classifier when they just weren’t. I’m absolutely shocked this is the first time this “type” of model has come to market.

These models will unlock AI utility for entire industries: robotics, trucking, aviation, meteorology, finance, defense, retail…even your smart coffee mug is going to use this.

3. It’s still too expensive

Jev is really, really cheap compared to LLMs. But you don’t use it like an LLM. In my full self-driving example, or any robotics example, you’ll be making decisions several times per second continuously. The #1 value proposition of a model like this is speed and cost.

To be clear, at current pricing, the model will still be successful and well integrated. It will be able to replace a number of tasks that were already being performed (poorly) by LLMs. However, to unlock industrial-scale demand (Jevons paradox), I believe it needs to be roughly 10x cheaper.

Currently, Jev is $0.042/M input, which is actually 7x more expensive than a cache read on DeepSeek V4.1 Flash ($0.006). If our industrial application requires decisions at 10 Hz and we provide only 10k of context, that would cost $0.0042/second to operate, or $15/hour, $600/week, etc.

The good news is Diogo said on a recent AMA that they have substantial margin at their current pricing and have already considered dropping the cost further. I hope it's a lot further.

4. Latency is a real limiter

I really hope TypeSafe AI works with an edge network provider to improve the co-location of models and reduce network latency. Ultimately, these models really need to run on-device. I would even be willing to pay some kind of recurring license fee to have a great closed model running on my own hardware, just so I can reduce the TTD (Time To Decision) as much as possible. Imagine running a model like this at 60 Hz, or even 120 Hz.

At that point, we’re looking at a new kind of logic gate — we can’t even imagine what that will be like.

In summary, this one really is a game changer for everyone building AI products, and, assuming cost and latency are further improved…it’ll be a game changer for everyone doing anything.

[Adam Thyer @AdamThyer]

@theo @braintrust @typesafeai Some people can't see ourside their existing paradigms. It's NOT an LLM! It's an ultra-fast ultra-cheap classifier that provides confidence scores and doesn't hallucinate. If you can't see the value in that, you lack imagination.

[Adam Thyer @AdamThyer]

@kunchenguid Because their default is to be a dimwitted next-token predicter. They're capable of much more, but most people don't know how to coax that out of them.

[Adam Thyer @AdamThyer]

@notkevinzhang @CompleteSkeptic @ainativefirm Was about to ask the same. I'll like the name 100x more if it is 🤣

[Adam Thyer @AdamThyer]

Food for thought 🤔♥️

[Michael Arnaldi @MichaelArnaldi]

@dok2001 Effect + Alchemy + DOs (and cf primitives) is an insanely powerful combo