Dean

Just made magic happen. @bloome_ai Apply For Me agent is finally working, backed by @browser_use. after matching you with really great job opportunities, with a single click, we will ship your application. Captcha, OTP email codes, authentication- all solved under the hood.

@mamagnus00 how can I try this? what should I set up?

We made browser agents 10× faster. First time your agent visits a website, it learns how. Next time, it just executes; no exploration.

@hcompany_ai this model is mind blowing! I'd love to use it in @browser_use .

Holo3 is here 🚀. Today, we're launching Holo3: our new series of frontier computer-use models. 78.9% on OSWorld-Verified. That puts us ahead of GPT-5.4 and Opus 4.6, at one-tenth of the cost. Weights on Hugging Face. API is live. Test it now! #Holo3 #OpenSource #ComputerUse #OSWorld #AI #AgenticAI

@hcompany_ai At Bloome, this is very relevant to how we think about job applications. Matching someone to the right role is important. Getting through the actual application flow is often the harder problem.

A lot of “agent” demos look impressive right up until the first weird popup, broken form, or verification step. Real value starts when models can deal with messy interfaces reliably, not just click through happy paths. That is why releases like Holo3 matter. @hcompany_ai #holo3

Holo3 is interesting for a reason that actually matters to builders. This is a reminder that the real bottleneck in agents is not language, it is execution. Anyone who has built browser/computer-use agents knows the hard parts are usually the same:

@LexnLin Both are terrible 😢

left: DeepThink right: GPT5.4 Pro Same prompt btw

@Homer_Wang @Tiny_Fish Great, keep me posted! Happy to continue over discord

@Tiny_Fish @_dean_sh That’s a great idea Dean. We’ll get this done for you:)

Give Claude Code, Codex, OpenClaw, Cline access to the entire web from your terminal. Introducing TinyFish CLI + Skills. 15+ agent workflows ready to use. CLI runs browser automations from bash for max token efficiency. Skills tell agents how to use that CLI. 500 free steps.

Getting agents to reliably handle messy application flows, weird websites, edge cases, and all the friction of the real internet has been one of the hardest things we’ve built so far. But that’s exactly why we’re so excited about it. More soon.

So now, when Bloome finds you a role that really fits, you’ll be able to click Apply For Me and let Bloome handle the boring parts for you. It fills things in, uploads your CV, moves through the flow, and only pulls you back in if needed. Simple idea. Very hard product.

Something big is coming to Bloome. We’re genuinely really excited about this one. We’re giving our agents a new superpower: Apply For Me. Finding the right role is one problem. Actually applying is another.

@Tiny_Fish Would be really ideal if via CLI Claude could see the logs of the run :)

@_dean_sh great question! claude would fall back to its native web search + fetch tools if it runs into issues with tinyfish api. and yes, you can see all the runs on your tinyfish dashboard!

@browser_use That sounds amazing. Can Claude access browser use logs, and correct for issues?

Supercharge Claude Code with Browser Use! 🍄 Try the Browser Use ecosystem in Claude Code > Browse the web like a human > Free, unlimited cloud browser hours Get started below ↓🔗

5) Store passwords in a secure vault. @Tiny_Fish just added 1Pass and bitwarden support!

4) cost matters. to drive down the amount of steps your agents execute, streamline the process for them to small, repeatable chunks: for example, "scan the webpage and extract all fields", then, "input {...} in this order". this way, your agents don't loop into uncertainty.

One of the biggest challenges I've had while building @bloome_ai Apply For Me was dealing with Anti Bot measures. (surprise, it's a bot). Here's a bit of helpful tips from my experience running on @hyperbrowser @browser_use and @Tiny_Fish 🧵