The interface is disappearing, and that changes who technology is built for
For most of the last thirty years, using technology meant clicking, tapping, or typing into something. You opened an app, found the right menu, and told the machine exactly what to do, one instruction at a time. That model is starting to bend. A newer class of software doesn’t wait to be told where to click — it’s given a goal in plain language and left to figure out the steps: book the flight, reconcile the spreadsheet, rewrite the onboarding email in a different tone. The screen hasn’t vanished, but it’s no longer the thing you’re negotiating with.
That shift changes who gets to define “easy to use.” When the interface was a grid of buttons, good design meant clear buttons. When the interface is a conversation, good design means the system asks the right follow-up question before it does something irreversible with your calendar, your inbox, or your money. OpenAI, Google, and Microsoft are each staking out a version of what that judgment should look like, and none of them agree yet on how much autonomy a piece of software should be trusted with before a person has to sign off. The company that gets that balance right first won’t just ship a better assistant — it’ll set the default expectations everyone else has to design against.
Why every major lab is suddenly obsessed with “reasoning”
For years, the fastest way to make an AI model better was simply to make it bigger — more parameters, more training data, more compute. That approach is running into diminishing returns, and it’s being joined by a different one: models that pause and work through a problem in intermediate steps before committing to an answer, closer to a person sketching out a proof on scratch paper than one blurting out a guess. The gains from this “thinking longer” approach have been large enough that Anthropic and its rivals have started measuring progress less by raw model size and more by how well a system reasons through multi-step tasks it wasn’t explicitly trained on — a shift that’s quietly changing what “state of the art” even means in AI.
The GPU shortage nobody saw coming twice
Manufacturers said supply would catch up to demand this cycle. It hasn’t. NVIDIA‘s order books stretch out for quarters, and the bottleneck isn’t the chip design anymore — it’s the handful of factories in the world capable of packaging it.
“We used to ask what a computer could do. Now we ask what it should be allowed to decide on its own.”
— a recurring line across this month’s AI policy hearings
Foldables finally stop feeling like a compromise
The first wave of folding phones asked you to accept a visible crease, a thicker body, and a hinge that gathered dust for the privilege of a bigger screen. That trade-off is disappearing. Newer hinge mechanisms from Samsung, paired with a widely rumored entry from Apple, suggest the category is finally being engineered as a primary phone rather than a showcase gadget.
Passwords are dying slower than promised
Passkeys were supposed to make “forgot your password” a relic within a couple of years. The technology works reliably where it’s implemented — sign-in is faster and phishing-resistant by design. The holdup has nothing to do with the cryptography and everything to do with incentives: most sites still treat passwords as the safe default and passkeys as an optional extra, so the habit that actually needs breaking is institutional, not technical.
Reusable rockets made this launch schedule boring — that’s the point
A launch used to be an event people gathered to watch. Now it’s closer to a freight schedule. SpaceX‘s reuse cadence has pushed orbit from spectacle toward infrastructure, and the companies that benefit most aren’t in the space industry at all — they’re the satellite-internet, imaging, and logistics businesses that now treat a rocket launch the way a shipping company treats a cargo run.
What regulators actually want from Big Tech this time
Earlier antitrust fights were mostly about scale — was a company too big, did it control too much market share. The current round of scrutiny is narrower and, in some ways, more consequential: who controls the flow of data between one company’s products and everyone else’s, and whether a platform can quietly favor its own services in ways a smaller competitor could never counter.
The quiet return of the tech IPO
After several years of well-funded companies choosing to stay private rather than face public-market scrutiny, that calculus is shifting. A new wave of infrastructure and AI-tooling companies — the unglamorous businesses that make other companies’ products work — is testing the public markets again, betting that steady revenue will matter more to investors right now than a dramatic growth story.
Inside the effort to make electric cars charge like a fill-up
The single biggest complaint about electric vehicles has never really been the driving — it’s the waiting. That gap is closing faster than the headlines suggest. Tesla opening its charging network to competing car brands has turned a fragmented patchwork of incompatible plugs into something closer to a shared standard, while newer battery chemistries out of Asian manufacturers are pushing charge times down toward the length of an actual fill-up. The remaining obstacle isn’t the car or the battery — it’s the electrical grid underneath the charging stations, which in many places was never built to deliver that much power that quickly, and upgrading it is a slower, less glamorous project than anything happening under the hood.
Also worth your time, elsewhere
