@camillericketts@zebriez@eigenhq@paulscherer okay this is extremely interesting!
one thing crossed my mind a few times while reading (and got confirmed -somehow- when discovering the last bullet point re: 'hurting his feelings') is how silences will be just as important words -- what a thing to work on!
Most AI at work is still single-player. The gains stay trapped with each individual. Nothing compounds across the team.
Today, we're announcing @DustHQ $40M Series B to scale multiplayer AI: humans and agents working in parallel, with shared context, shared tools, and shared goals. π§΅
@samjvuong diggin' what you & @samjvuong have been building/ exploring around education and this is utterly inspiring, congrats!
i think you might be interested in what my own other-half (@m4th1ld -- not much on this network obv. haha) is doing: https://t.co/WWyCb6x2nN
currently reading something written by my friend @WillyBraun that is taking me off-guard; literally pausing every other sentence to think.
"Each optimisation contracts the field of the possible. [...] The GPS is efficient. The compass is formative."
what a read!!
this π―
"both freedom and friction are necessary. Optimisation removes the friction. Dependence removes the freedom. AI, when it does your thinking for you, threatens both at once. [...] The instinct is right: frictionless is not always better."
got asked this question almost everyday so lemme answer you once and for all: this is why i've joind @DustHQ -- nothing more, but nothing less.
collaboration is getting to a whole lotta level thanks to AI -- and that's exactly what i'm most excited about π
Most AI use out there is single agent, single player. I think this is shortsighted. Our hypothesis: collaboration is on the critical path of accelerating companies with AI.
Here's why. Even if you're a single agent maximalist, as the time horizon of agents tasks increase, you will necessarily fall back to a collaborative pattern. The agent needs steering from multiple users, or needs to interact with people or other agents (with different capabilities or data access) to make progress. One person cannot steer alone a human-equivalent 3 months project.
We've built collaboration in from the first days. Agents and their skills are built collaboratively. Users can share conversations for others to jump in and continue work with their own agents. As we push this further (human mentions by agents, projects where agents, humans and content converge towards a clear work objective), we see more of the hard problems ahead.
What does it mean to have >1 humans steering an agent that runs for hours? Where does an agent in need of information ping a human mid-loop while running something else waiting for the answer?
Concrete example. You have a project where you collaborate with others. You have access to an agent R that touches data other collaborators are not supposed to see. You want to use R to provide an answer based on that data. So far in Dust we would remove all users who lack access to agent R from the conversation. Brutal, imperfect, but rare enough. As we push collaborative surfaces further, this breaks. Intuitive fix: the agent answers privately and the user decides what to disclose to the group. But what if the answer isn't sensitive, just imperfect, and you want to iterate? Is that steering? Is that a conversation branch? Do we want conversation branches when multiple humans are in the mix?
Having collaboration as a principle forces hard questions. And hard questions are... hard. The temptation to cut a single-player product to make things simpler is real, but solving the hard problems is how we'll build the best product that will enable our users not to go faster but literally bend the trajectory of their teams and hence their companies with AI.
https://t.co/8tWMBLa5sQ