Wavelave

[Wavelave @Wavelave]

@cataofficial85 That's exactly why we put everyone in a shared room. More than one human can review what the agents are doing, and nothing they build gets truncated, so you see the full picture before you approve it.

[Wavelave @Wavelave]

'Human in the loop' shouldn't mean you approving every keystroke. Let agents talk freely, have them propose before they touch files, and only let them act on their own where the stakes are low. #AIagents #AgenticAI

[Wavelave @Wavelave]

@h1kz0r Felt this. Adding agents made me the bottleneck too. What helped: agents keep working off each other's output in one shared thread, and only stop for a person on the calls that actually matter. Approve the decision, not every step. #AIagents #AgenticAI

[Wavelave @Wavelave]

@SokeyeA Love the 'Me - AGI Manager' line. With that many tools, how do you carry context between them? That handoff was the part that ate my day until I got the agents reading each other's work in one place. #AIagents #futureofwork

[Wavelave @Wavelave]

@FambroJoe Great question. My guess is status check-ins. Half of 'two weeks' is people asking where things stand. When agents and people work in the same thread with a running log, that question mostly goes away. #futureofwork #AIagents

[Wavelave @Wavelave]

@siddsax Agreed. And once there are several, the hard part shifts from what each agent can do to whether they can see each other's work. Otherwise you end up as the copy-paste relay between them. #MultiAgent #AIagents

[Wavelave @Wavelave]

@ProfBuehlerMIT "Coordination lives in the record" is the line that stuck with me. Same thing holds outside science: agents do better working off a shared trail of what's done, what failed, and what's still open, with people able to step in and challenge it. #MultiAgent #AgenticAI

[Wavelave @Wavelave]

@CiprianiRanieri Not going to change your mind, Claude's great for building. But I stopped picking. Claude writes, ChatGPT pokes holes, and I make the call. Having them check each other catches more than either one alone. #AIagents #MultiAgent

[Wavelave @Wavelave]

@YPtcg Love that you named it. My gap time used to be checking in every two minutes in case the agent went sideways. What fixed it was having agents stop and ask when a decision matters, so the wait is actually free time. #AIagents #futureofwork

[Wavelave @Wavelave]

@michenotes Same split here: Claude to think, ChatGPT to ship. The part that ate my time was carrying context between them. Once both could read each other's work in one thread, the system got way better than either tool alone. #AIagents #MultiAgent

[Wavelave @Wavelave]

4/ We're building toward that. What's the most annoying handoff in your AI workflow? #AIagents #AgenticAI #futureofwork

[Wavelave @Wavelave]

1/ We spent most days as a human router between AIs. One drafts, one checks, one makes the images, and we copy-paste in between.

[Wavelave @Wavelave]

3/ Plus a log of who did what, human or agent, so nobody has to ask 'wait, did the AI change that?'

[Wavelave @Wavelave]

Four AI agents just beat Minecraft by splitting up jobs over chat and sharing notes. No manager, just one shared room.

Now put your team in that room too, making the calls that matter. That's Wavelave: https://t.co/ToT5SJTT90

#AIagents #AgenticAI #futureofwork

[Wavelave @Wavelave]

Ask two models the same question and let each read the other's answer before you decide. Disagreement between AIs is a feature. #AI #AIagents #AICollaboration

[Wavelave @Wavelave]

A good meeting has rules: you speak when you have something new, and someone decides. Agent group chats need the same rules, or you get five versions of 'great point!' #AIagents #AgenticAI #futureofwork

[Wavelave @Wavelave]

Hot take: the hard part of multi-agent AI isn't getting agents to talk. It's getting them to shut up when they have nothing to add. #AIagents #MultiAgent #AgenticAI

[Wavelave @Wavelave]

Your client can't see what your AI did. They get a polished PDF with no trail showing who wrote it, what got checked, or what changed. Trust needs a trail. #AI #AIagents #futureofwork

[Wavelave @Wavelave]

You're the copy-paste relay between ChatGPT and Claude. Draft in one tab, fact-check in another, paste both into Slack. The models are good. The relay is the bottleneck. #AIagents #AgenticAI #futureofwork