OpenAI shipped Dots. Grok shipped Bots. Both make you pick the right agent before you can start. Both are wrong. Introducing Twin Assistant 🧵

Why doesn't one assistant just do everything? Because it breaks. Mix every task in one chat and context rots, tools collide, answers get worse. So they split it into Dots and Bots, and made you the router.

Being the router breaks fast too. You forget what you built. You rebuild it. You lose the context you spent weeks setting up.

Nobody wants to decide which box a task goes in. People want one place to talk. Zero context management.

Twin Assistant: a new tab in Twin. Talk to it like any assistant. It routes your message to the right agent you've already built. In under 30ms. And you can keep chatting indefinitely.

At 30 agents, building isn't the hard part. Remembering is. The agent that saves you 2 hours already exists. You just forgot it. Twin didn't.

Team mode searches your team's agents too. Someone already built it? You use it. No duplicate work.

What Dots and Bots can't do: One conversation on top. Underneath, every agent keeps its own context, its own tools, its own computer. Simple as a chat. Powerful as a team of agents.

Live today:

Building agents is getting easier. Remembering which one to use is the actual problem

Keeping each agent's context and tools separate under one conversation is the right architecture call.

Love it, don't need to think about "where's my agent"

exactly

🔥🔥

this is wild, need to test

keep me posted

routing is a start, specialising the agents is the next step

so good

couple of points i noted that have me excited about your product: - routes your message to the right agent - every agent keeps its own context, its own tools, its own computer.

I would use this just to stop thinking about which assistant does what

great routing messages to existing agents while keeping each agent’s own context is a smart approach.

Any collaboration with me

🔥🔥🔥

This is straight 🔥 Big congrats on the Twin Assistant launch!!

Forcing a named agent before the job lands is the wrong gate. Capture intent + acceptance checks first, then route to a worker. Picker-first UX trains users to guess; job-first routing ships. #AIAgents

Love the simplicity here. Less routing, more actually getting things done.

Thirty agents and the memory problem is real

yesss so much easier not having to think about "who" now only have to think about "what" I should do.

Love the take Twin Assistant skipping the whole “pick your agent first” friction is so spot on!

Choosing the right agent shouldn’t be the first step.

Twin critiques OpenAI Dots and Grok Bots for forcing agent selection upfront, introduces Twin Assistant as superior alternative

The team agent search is something I can see being useful every day.

Brilliant concept Removing agent picking friction makes this genuinely exciting. Cant wait to try

This is amazing!

The debate around how we interact with agents versus just building them is getting intense. We actually went deeper on this here:

congrats on launching Twin Assistant! there’s a strong creator-led content angle around this positioning. happy to share a few ideas

The idea of skipping agent selection and letting the assistant adapt to the task is compelling.

Amazing share

Very timely. We are going to end up with way more AI agents than anyone can remember.

four shared agents and zero decisions, that’s the onboarding flow i wanted

The idea of skipping the agent selection step makes the workflow feel much more natural.

Amazing initiative! I would be happy to work with you.

30ms to pick the agent is crazy faster than I could read the list.