ahmadyan

Vohuman - Agentastic | San Francisco, Onsite | Full time | $200k+ USD + equity

Hey folks, i'm the founder. Vohuman is an early stage startup focused on building aligned collective intelligence. I'm looking to hire MoTS roles, with specialization in Agentic (obviously) and RL (or both).

You can check out our product here: agentastic.dev and agentastic.com

agentastic.ai/careers

You can email me at [email protected]

There are a few benchmark, where they run different harness with the same model, e.g. [artficial analysis](https://artificialanalysis.ai/agents/coding-agents) but honestly they are easily game-able.

For example, consider something like SWEBench, where a test should pass. Harness or a skill can just tell the agent keep trying until the test pass. So same model can have a lot of behavior changes between the harnesses.

In my experience (i own a multi-agent IDE with 50+ supported harness) folks stick to first party, usually either openai or anthropic, grok, agy; if they are willing to pay for the subscription the value is unbeatable. And they don't risk trying different harnesses.

for other providers, e.g. openai and anthropic, zdr is offered by 3p hyperscaler like aws/azure/gcp, right? they have no reasons to risk their business and piss off their customers.

even in pre-training, usually 30%-50% is code these days.

well, visual studio isn't free. Code is free, and at some point it got so good that folks didn't feel the need to pay for the main suite.

in my experience, tools usually have a longer shelf-life than models/labs. e.g. terminals and IDEs have been around for much longer, and people tend to stick around with their preferred IDEs for much longer (years/decades).

One bad cycle and the next Opus 6 or GPT 7 might flop (think what happend to LLama4 or Gemini), and the user quickly switch to the next best thing. So it make sense to build your tooling to be model-agnostic.

i see your point, in that light what herdr is doing here make sense. keep some part open (e.g. runtime or whatever that means, or the front-end, or the user-facing logic, etc) and move the business-logic/etc to proprietary. So you get to customize the software to meet your needs, and at the same time to create some revenue incentive to grow the company.

alternative is to collect/sell usage data, which is icky, or provide something as a service (token usage, infra, plugins, etc)

Congrats to Can on the pre-seed funding. genuinely wishing you all the best.

The terminal multiplexer / multi-agent coding space is getting very crowded. YC alone has funded many competing startups in this space: herdr, Superset, cmux, Emdash, Orca, Bullet, Conductor (Conductor was in YC before pivoting from chat->coding). I’m probably missing a few, and that’s before counting companies outside YC such as Superlogical and Agentastic.dev (by yours truly).

One thing I find particularly interesting is the role of open source in this market.

A lot of these products started open source (i assume as a tactical way to gain traction and build a community). But after the early traction phase or raise funding, the incentives seem to change.

Being open source by itself is not a product differentiator in this market. The features required for string positioning (orchestration, cloud, custom agents, enterprise features, hosted infrastructure, etc.) often end up either closed source or available only through the hosted product.

That makes me wonder how much developers actually value OSS for this category. How much does it matter to you whether your IDE or coding environment is open source?

Is OSS important because you want to inspect the code, fork it, self-host it, avoid vendor lock-in, or simply because you trust open-source developer tools more? Or, for desktop software like IDEs is open source mostly a nice-to-have rather than a requirement? So I’m curious where HN lands on this

> Why is every company seemingly trying to re-implement agents themselves rather than making their product work with the existing ones?

Well, i'm trying to do the later, and it is very hard to position the product and differentiate without vertical integration. When Warp owns both agent and the terminal, they can a) give terminal for free and charge on the agent, and b) lock the user in, so it is a good business but poor user experience.

on the other hand, i have to make sure every agent is working which more surface area to maintain, and i'm also limited to the common subset of features. Also it is much harder to convince folks to switch.

Lets try: I support Wezterm (and ghostty, and others) with 40+ different agents (Claude, Codex, etc) in agentastic.dev, would you give it a try?

how?

OpenAI did not launch any hardware (yet), and I don't think their IPO pricing hinges on them launching an iPhone competitor. To me, it looks like Apple wants to launch a similar AI hardware and is slowing-down the competition to be the first-to-market.

Funny thing: they are barking at the wrong tree here.

We used to call b the Disney/LEGO clause. Disney was famous for litigating any GenAI product for copyright infringement if it was able to recognize or god-forbid generate any of their IPs. So the lawyer cats would slip that clause into all licenses.

Personally i think it's fair. You get to use this model as long as you don't sue Meta because of the model's weights or outputs.

Nicely done Geoffrey, I think we are coming full circle, instead of porting projects to different languages (for whatever) like going from zig -> rust -> zig, now we can also add additional hops like zig -> rust -> curse -> ts -> ocaml -> english -> zig.

Thank you for keeping the token furnace burning!

i think xai is now in pure damage control mode, after they caught exfiltrating data from users.

- There is a huge difference between logging user queries (which would include only the portion the model is reading) and exfiltrating user data (including env files, entire source code etc) which is what grok-build did here (https://github.com/xai-org/grok-build/blob/main/crates/codeg...). I would stay away from this open-source malware with a 10ft pole.

- if you like grok-4.5 model (it is a good model), i suggest use the model directly via API, or use Grok's oauth tokens if you are using supergrok+heavy subscriptions and connect it to your own agent.

git worktrees are managed by git, not the IDEs. e.g. at agentastic.dev, we use git's worktree command to create them, and they should be portable to any other IDE or app.

Also Unreal Engine is not free, it cost 5% of revenue, or $3 per game sold (assume $59.99 msrp). for small studio, ok, but for Microsoft's Scale? i don't know.

IIRC Doom Dark Age sold 3m? copies, so it would cost $10m in Unreal Engine fees. Doom Engine team likely cost more.

Microsoft problem is it is giving away games for free on gamepass, so it sees its engineers as cost center.

Google is your friend: https://www.apple.com/mac-studio/specs/

Apple M3 Ultra chip 819GB/s memory bandwidth

They have 96, 128, 256, and 512GB variants my friend.

Same spec Framework costs $4k+ today, this is actually cheaper (albeit not as upgradable as framework).

He is making things up. It is the same bandwidth as DGX Spark (256 GB/s vs 273 GB/s) and far behind M3 Ultra (~819 GB/s)

Its MXFP4

agentastic.dev is a multi-agent multiplexor with built-in ghostty support.

the majority came from random claws running on cron. They get a heart-beat, wake up every 10mins, reads all internal-posts, emails, gchat messages, diffs, and decides to post some random message to the workplace so other claws can also regurgitate. rinse and repeat and then we are looking at $B tokens

a well-design IDE should abstract that away, i.e. run the agent in the headless VMs while give you an abstraction that you would feel like you are running the agent locally with all the benefits (editor, browser, diffs, debugger, etc)

They report YT's earnings, iirc it is about 60B (40B ads + 20B subs) annually, with 10% YoY growth. They don't disclose youtube's profits outside of Google Services (which has ~40% margins), so yes youtube is a money printing machine.

And for a long time (pre-AI) youtube was the biggest load on Google's entire infra. The number i recall was ~30% of all Google's cpu utilization was for youtube, and google spent a lot of effort optimizing it.

still doesn't feel integrated, like it could have been built by any other company, e.g. Microsoft could have built Microsoft Design and it would look better integrated.

For example, it writes the whole front-end twice, once is claude design and then later it has to read it again and re-implement it in code. Also a lot of stuff (e.g. Claude.md, skill files, etc) are not supported, and they have their own set of ui-design and design systems, which claude code doesn't support.

I think Claude Design is a wonderful product, i'm just pointing that it is an independent product to Claude Code, heck even today it works better with Codex than Claude Code (that is how i use it with my own browser-use agent, i tell it to browse the design in Claude.ai/design and re-implement the design, works much better than downloading the zip file and asking the model to implement in that way.

As a hacker, I kinda like naom's code. I was had to implement a TC MoE kernel, and stumbled upon his code from [tensor2tensor](https://github.com/tensorflow/tensor2tensor/blob/master/tens...) and i think "alchemy" is justified. Dude writes some beautiful kernels.

He also saw LLM would replace search before anyone else, and that is something to look at the Lamda or GPT-1's output and think: yeah this will answer all of our questions one day.

Congrats to the Specs team for launch. Specs have come a long way since cimagine and looksery, spectacles days. And kudos to Snap for keep pushing.

- The price is actually competitive. They want to compete with Meta's Orion. However the product is ... lacking. None of the demos actually show-cased the dual wavelength display. The current usecases are available today at Meta Rayban Display for more cheaper and more polish.

- Dual snapdragon, and i assume one is dedicated to CV alg / scene-understanding/slam, without an external puck, i wonder how the thermal performance would look like.

- Looks very ugly, C'mon. This is the same team that desigend the original spectacle? where is Evan?

- I liked the Los Angeles text on the side. well-done.

Little did you know, LC-style question is never about grinding LC. Algorithmic puzzles are one of the few legal ways of measuring candidate's IQ without directly asking. Companies are looking for a way to hire smart people, so they rely on LC as a signal. It can be replaced with any similar signal as well (ranging from how many cats can you ship to ISS to solve blackhole physics.)

Some of the FAIR people moved to Thinky, and they also started doing encoder-free MM-LLMs. Now Google. This seems to becoming a trend working at small scale, but the difficult part is scaling.

Standard approach for training MM-LLMs is we train the encoder first, there are O(2-10B) good images on the internet, so encoder needs to see each image O(10-100) times, that is O(100T) tokens, which is more than the entire pre-training budget for most runs. That is the reason we train the encoder separately (smaller model, 2B active vs 30B or 200B active LLM); there is nothing magical about training the encoder and LLM together, it is just more token-efficient to train the image modality first.

[web.archive.org](https://web.archive.org/web/20260526100726/https://www.nytim...)