Pine Computer is a cloud computer designed to be operated by AI agents rather than by people, launching in private beta for developers through an SDK. A developer's product creates a Pine Computer when a job needs one and hands it the job. Pine Computer carries the work itself, across websites, files and business software, and the finished work comes back into the product. The pitch to builders of agent products is concrete: instead of driving a human desktop through screenshots and mouse clicks, Pine reads a web page as structure (what is on it, what can be done with it, what just changed) rather than reconstructing it from a picture each time.
The launch matters now because agent builders are hitting a ceiling on speed, cost and reliability when wiring models into browsers and SaaS tools. Each Pine Computer runs sealed in its own sandbox, and the developer's keys stay with them. A live screen can sit inside the product, so its users can step in when a task needs them. Developers who want to bring their own model can connect it too . In Pine's tests, Pine Computer is 2, 5× faster than AI on conventional computers. On UniPat AI's SaaS-Bench v1.1, a public benchmark of 106 business workflows across 23 applications, it has the highest checkpoint score in the publisher's table: 78.3%, against 74.3% for Opus 5 with Claude Code and 71.1% for GPT-5.6 Sol with Codex. It ran GPT-5.6 Luna, a lower-cost model, at about $1.02 per task in model cost, against $26.50 and $20.50 for those systems.
Pine AI, the team behind the launch, was founded by Stanley Wei and Vincent Sun, former colleagues at Agora Inc. (Nasdaq: API), where Stanley held the role of CSO & COO, and Vincent served as VP of Engineering . The product is being introduced publicly by Stanley Wei on X and by Dylan Wang, Pine AI's co-founder and chief architect , who frames the project as open by intent, saying Pine plans to open its designs, implementations and specifications, and to work with others on what a computer built for AI should be. Pine AI already runs the system inside its own consumer and enterprise assistants, and access for outside developers starts with a waitlist at pinecomputer.io.
Comments (13)
Sign in to join the discussion.
N
Nkechi E.2d ago
The tweet opens with 'AI is already smart enough' which is the only hook that could make me stop scrolling at this hour. Rest of the thread does the heavy lifting though, that first line is doing cardio.
Y
Yannis Petrou2d ago
Suggested rewrite, free of charge: 'Your AI is driving a car designed for feet.' You can venmo me later.
R
Ravi K.2d ago
Building in the agent runtime space too and this framing is annoyingly good. I was going to use 'harness' in my own copy next week, now I can't.
E
Elin Lindqvist2d ago
What's actually under the hood here? Is this a new VM abstraction, a kernel-level thing, or a browser-shaped sandbox with better affordances for tool calls?
J
jules the intern2d ago
naive q but if the computer is built for AI, how do i, a human, see what it did? like is there a screen or
T
Thabo M.2d ago
Docs link or it didn't ship. Also curious if any of the runtime is open or if we're all just trusting the pine.
L
late adopter larry2d ago
Call me when it runs Excel. I'm still waiting for the Zapier integration on half the 2022 launches I bookmarked.
Y
Yilin Chen2d ago
The launch video pacing is weirdly calm for an AI infra reveal, no synthwave, no fake terminal typing. Respect for not doing the Vercel cosplay.
K
Kira Novak2d ago
Genuinely how many people built this? The surface area implied by 'a computer for AI' is bananas for a seed-stage team.
M
Mikhail D.2d ago
Love the vision, now tell me how it's provisioned. Do I spin one up per agent, pool them, or is there a weird cold-start story nobody wants to talk about.
A
Amara Osei2d ago
Opened the site expecting a 'get started' button in the first viewport, got a manifesto instead. Beautiful manifesto, but my cursor is lonely.
D
demidov2d ago
Every 'built for AI' demo is a scripted happy path until I can paste in my messy real workflow. Will reserve judgment until something breaks on camera.
D
Dr. Halvarsson2d ago
The 'computer built for humans' framing echoes the embodiment arguments from the HCI literature, except now the body is a transformer. Would read a whitepaper if one exists.
Work With Us
Book an intro call to explore our suite of services. We might have a waitlist.
Pine
A computer built for AI
A computer built for AI
Tags
TLVC Rating
Community Rating
No ratings yet
to rate this launch.
Pine Computer is a cloud computer designed to be operated by AI agents rather than by people, launching in private beta for developers through an SDK. A developer's product creates a Pine Computer when a job needs one and hands it the job. Pine Computer carries the work itself, across websites, files and business software, and the finished work comes back into the product. The pitch to builders of agent products is concrete: instead of driving a human desktop through screenshots and mouse clicks, Pine reads a web page as structure (what is on it, what can be done with it, what just changed) rather than reconstructing it from a picture each time. The launch matters now because agent builders are hitting a ceiling on speed, cost and reliability when wiring models into browsers and SaaS tools. Each Pine Computer runs sealed in its own sandbox, and the developer's keys stay with them. A live screen can sit inside the product, so its users can step in when a task needs them. Developers who want to bring their own model can connect it too . In Pine's tests, Pine Computer is 2, 5× faster than AI on conventional computers. On UniPat AI's SaaS-Bench v1.1, a public benchmark of 106 business workflows across 23 applications, it has the highest checkpoint score in the publisher's table: 78.3%, against 74.3% for Opus 5 with Claude Code and 71.1% for GPT-5.6 Sol with Codex. It ran GPT-5.6 Luna, a lower-cost model, at about $1.02 per task in model cost, against $26.50 and $20.50 for those systems. Pine AI, the team behind the launch, was founded by Stanley Wei and Vincent Sun, former colleagues at Agora Inc. (Nasdaq: API), where Stanley held the role of CSO & COO, and Vincent served as VP of Engineering . The product is being introduced publicly by Stanley Wei on X and by Dylan Wang, Pine AI's co-founder and chief architect , who frames the project as open by intent, saying Pine plans to open its designs, implementations and specifications, and to work with others on what a computer built for AI should be. Pine AI already runs the system inside its own consumer and enterprise assistants, and access for outside developers starts with a waitlist at pinecomputer.io.
Comments (13)
The tweet opens with 'AI is already smart enough' which is the only hook that could make me stop scrolling at this hour. Rest of the thread does the heavy lifting though, that first line is doing cardio.
Suggested rewrite, free of charge: 'Your AI is driving a car designed for feet.' You can venmo me later.
Building in the agent runtime space too and this framing is annoyingly good. I was going to use 'harness' in my own copy next week, now I can't.
What's actually under the hood here? Is this a new VM abstraction, a kernel-level thing, or a browser-shaped sandbox with better affordances for tool calls?
naive q but if the computer is built for AI, how do i, a human, see what it did? like is there a screen or
Docs link or it didn't ship. Also curious if any of the runtime is open or if we're all just trusting the pine.
Call me when it runs Excel. I'm still waiting for the Zapier integration on half the 2022 launches I bookmarked.
The launch video pacing is weirdly calm for an AI infra reveal, no synthwave, no fake terminal typing. Respect for not doing the Vercel cosplay.
Genuinely how many people built this? The surface area implied by 'a computer for AI' is bananas for a seed-stage team.
Love the vision, now tell me how it's provisioned. Do I spin one up per agent, pool them, or is there a weird cold-start story nobody wants to talk about.
Opened the site expecting a 'get started' button in the first viewport, got a manifesto instead. Beautiful manifesto, but my cursor is lonely.
Every 'built for AI' demo is a scripted happy path until I can paste in my messy real workflow. Will reserve judgment until something breaks on camera.
The 'computer built for humans' framing echoes the embodiment arguments from the HCI literature, except now the body is a transformer. Would read a whitepaper if one exists.