Bluesky launched group chats on Thursday and outlined a broader pivot toward community features, a strategic shift for a social network that has so far focused on open public posting. Group chats support up to 50 people in the current release. The company says it may increase that limit later.

The feature arrives in version 1.124 of the app. Chat creators control who can participate and can generate invite links that display as embedded cards when shared in Bluesky posts. Users can set who is allowed to invite them: everyone, only people they follow, or no one. Media sharing in group chats is not yet supported because it requires additional safety and moderation systems.

The more significant announcement is what comes next. Head of product Alex Benzer said Bluesky will build communities, smaller spaces inside the platform where users go deeper on shared interests. Communities will have their own handle that doubles as a URL, like community-name.bsky.social. They can be public, invite-only, or private.

“Today, Bluesky is one big space. Communities will be smaller spaces inside that where you can go deeper and hang out with people who care about the same stuff,” Benzer wrote. The features will be built on the underlying AT Protocol with support from the wider developer ecosystem.

The timing is deliberate. X shut down its own Communities feature in April, citing low usage and spam. Bluesky is picking up where X left off, targeting users who want more control over their online communities without Big Tech intermediaries. Bluesky’s Attie app already lets users customise their feeds with AI. Communities extend that philosophy to group interaction.

The pivot reflects a growth problem. Bluesky has 44.8 million registered users. X has 600 million monthly actives. Threads has hundreds of millions more. If Bluesky cannot match those numbers, it needs a different value proposition. Communities and group chats offer something X and Threads do not: user-owned spaces on an open protocol, where moderation rules are set by the community rather than the platform.

X recently moved in the opposite direction, launching a standalone XChat messaging app while shutting down its group community features. Bluesky is betting that small, controlled spaces matter more than massive public reach, especially for users who have already decided they want an alternative to platforms operated by Musk or Meta.

Whether 44.8 million users are enough to sustain vibrant communities is the open question. Reddit works because communities have critical mass. Bluesky’s total user base is smaller than many individual subreddits. The group chat launch and the communities roadmap are an acknowledgment that growth alone is not working, and that depth of engagement may matter more than breadth.

Impulse Space, a startup founded by SpaceX engine guru Tom Mueller to build highly-maneuverable spacecraft, announced a $500 million Series D this week that it will use to hire as many as 200 new employees.

The round, led by 137 Ventures and BANNER VC, with participation from Founders Fund, Lux Capital, and Linse Capital, reflects investor interest in space and defense tech as the U.S. government hurls cash at national security problems and SpaceX gears up for its IPO.

Impulse is focused on in-space mobility. The company has developed a highly maneuverable platform called Mira that is targeted at U.S. Space Force buyers. It’s also building Helios, a vehicle designed to carry satellites rapidly to high orbits after they are dropped off in space closer to Earth.

President and COO Eric Romo told TechCrunch that the new capital will help the company build and test more space vehicles and emphasized the company’s hiring plans at a time when aerospace talent is in high demand.

While the company’s software teams are adopting AI coding tools, Romo said that when it comes to solving engineering problems in the real world, deep learning models aren’t quite ready for prime time. As the 13th employee at SpaceX back in 2003, Romo’s job was creating computer simulations of the company’s engine design to assess its performance.

“I considered it success if I got within 20% of the right answer, because the simulations were just not that good,” Romo said. “They’ve improved, but they’ve not improved that much, and so there’s not really any substitute for designing the thing, analyzing the thing, building it, and then getting it on the test stand.”

Romo suspects AI tools for hardware design may be slower to arrive because the right training data is hard to find, compared to the amount of text and code available on the internet to train LLMs. “If you want to go, say, find the best designs for a turbo pump seal package in the world, you’re not going to find those online,” he points out.

Impulse started with a focus on propulsion and evolved to build spacecraft, requiring the company to add more expertise in the form of engineers who build vehicle structures and flight computers. One reason the company recently opened an office in Colorado is that aerospace talent has more options today — instead of just going to Los Angeles, engineers can find work in Seattle, Denver, or Texas.

Next up for the company is another launch of its Mira spacecraft, which made its third flight late last year. That flight wasn’t without incident — a problem with its navigation system led it to expend much of its propellant early on. Romo said the company is prepping a new Mira mission that is expected to launch before the end of the year.

As enterprises troubleshoot their AI systems, governance has emerged as a key challenge. Some are taking a dual approach: One model to handle incoming queries, while a second model keeps the first one from getting into trouble.

That’s the premise of ZeroDrift, a new AI compliance service that announced a $10 million seed round on Tuesday. (Investors include a16z Speedrun, Reign Ventures, PitchDrive Ventures, and U&I Ventures, among others.) The company deals entirely with the second part of the system, sitting between AI models and end users to flag and replace any messages that might present a compliance problem.

It might seem strange to build an AI system to correct other AI systems’ mistakes — but ZeroDrift’s correction system has a few architectural advantages over the models it will be correcting. The system is triggered by conventional programs that deterministically apply known compliance standards like SOC 2 or GDPR. The LLM only comes in once a message has been flagged, rewriting a compliant version of the same message.

“We’re able to identify deterministically, what are all the regulated areas, what’s the violation that’s being broken, and then we have LLMs that can do the rewrites,” Aroomoogan says.

Critically, the whole system can be run with less latency and more reliability than a conventional LLM. It’s the company’s primary advantage over big labs like OpenAI and Anthropic, which are often already present in the underlying system.

The most obvious use case is for AI chatbots, which are already deployed in front of consumers where there can be serious consequences for rogue answers. But Aroomoogan sees the total addressable market as much larger, potentially spanning AI-generated messages that human beings never see, generated only within automated systems. So far, it’s a relatively small market — but it’s one that will grow as AI proliferates.

If the fundraise is any indication, there’s a lot of pent-up demand for that product. “It was probably the fastest fundraising I’ve done in my life,” CEO Kumesh Aroomoogan says, crediting Andressen Horowitz’s help structuring the seed round. “We closed within three weeks, and we will be oversubscribed by 3x on the amount.”