Overview
Today’s thread runs on two tracks: the space story getting bigger and louder, and the AI tooling story getting more practical and more expensive. There’s also a quieter note on health data and risk, plus a reminder that markets and sceptics both love moving the goalposts.
The big picture
The common theme is scale. SpaceX is talking launch windows and IPO scale in the same breath as jaw-dropping engine photos, while NASA is pointing to the less flashy payoff of microgravity research that ends up back on Earth. On the AI side, the tools keep landing inside the software people already use, but the bill is starting to matter, and new ideas about paying for content are turning into products.
ISS science in 2025, practical wins from microgravity
NASA’s recap is a good reminder that the International Space Station isn’t just a backdrop for astronaut selfies. The headline items are surprisingly grounded: robotic surgery experiments that matter for remote medicine, and battery tests that speak to safety and longevity rather than hype.
The more sci-fi sounding bits, like magnetic levitation helping synthetic bone grafts form better structures, still point to a simple logic: change the environment, learn something you cannot learn on Earth, then bring it home.
Starship test day, with the usual sense of occasion
SpaceX flagged it was on track for a Starship flight test with a tight evening window, plus coverage starting ahead of time. It reads like a routine ops note, but the engagement shows how “routine” has changed for large-scale rockets.
A single photo that explains why people keep watching rockets
Sometimes the update that travels furthest is not a schedule or a spec, it’s an image that makes the hardware feel unreal. tetsuo dropped a shot of Raptor engines at full chat, and it hit that sweet spot between engineering and theatre.
If you want a quick sense of why Starship content keeps leaking out of space Twitter into everyone else’s feed, it’s this.
SpaceX IPO talk, and Goldman’s taste for uncomfortable wins
Trung Phan says Goldman Sachs is lead underwriter for the SpaceX IPO, then pulls a neat historical rhyme with Tesla’s 2010 listing. The point is less “bankers are brave” and more that big institutions sometimes need a senior person willing to take the internal argument and wear it.
It’s also a reminder that the same names keep showing up when a new category goes from odd to inevitable.
The web wasn’t built for agents, so people are rebuilding the plumbing
Firecrawl’s pitch is blunt: websites are for humans, agents need something else. Their /search, /scrape, and /interact framing captures where things are going, not just reading pages, but clicking through them, handling pagination, and dealing with logins.
If agent workflows keep growing, the winners will not just be models, but the boring-looking middleware that turns the messy web into usable context.
Grok inside OpenCode, subscriptions turning into model access
xAI is plugging Grok and X Premium into OpenCode, so a consumer subscription can become a developer tool credential. That line keeps getting thinner: the “chat app” identity is fading, and what’s left is model access plus decent integration points.
The important bit is not the announcement itself, it’s the direction: more coding agents that live in the terminal or desktop, with authentication and model choice handled like any other service.
ChatGPT in PowerPoint, the slow takeover of office work
Greg Brockman’s “ChatGPT for PowerPoint” post is short, but the intent is clear: put the assistant inside the tool where the work already happens. Generating slides is the obvious part, but querying and editing across a whole deck is the bit that changes how people maintain documents over time.
It is also a quiet signal of where these assistants are headed next: not more chat, more control surfaces.
AI coding costs are going to bite, so ship while it’s cheap
Bojan Tunguz has the warning developers keep repeating in private: build now, because today’s pricing might be temporary. As usage becomes normal inside companies, the spend becomes visible, and procurement starts asking hard questions.
Whether the next step is stricter limits, token billing, or internal replacements, the takeaway is the same: treat cheap capacity like a short-lived window.
Micropayments did not work for humans, but agents might change that
Patrick Collison points at a path that sounds obvious in hindsight: if the friction is the human decision, let agents do the deciding. Paying creators and publishers in small increments has failed for years because it asks people to stop, think, and choose, repeatedly.
Agents can make those choices automatically, which turns “tiny payments” from a headache into a background process. If that happens at scale, it could rewire how content gets priced and paid for.
Health metrics meet imaging, and the risk debate keeps going
jack posted his first CT scan results with and without contrast, showing no plaque and no stenosis. It is the kind of data that both reassures and provokes argument, because it sits awkwardly next to biomarker chat about LDL and ApoB.
The replies are the familiar split: measure what matters directly versus trust the proxy numbers. The useful thing is seeing the conversation happen around an actual scan report, not just theory.
