My own experience has been that you really just have to be diligent about clearing your cache between tasks, establishing a protocol for research/planning, and for especially complicated implementations reading line-by-line what the system is thinking and interrupting the moment it seems to be going bad.
If it's really far off the mark, revert back to where you originally sent the prompt and try to steer it more, if it's starting to hesitate you can usually correct it without starting over.
Was curious if ground vehicles at airports also use transponders to communicate position to the radio tower, and it turns out the FAA put out a report last year on potential solutions to avoid this exact situation:
My concern is less total link loss than what happens under degraded or intermittent connectivity. If the wingman still depends on the manned aircraft for tasking or weapons authority, then the interesting question is how it behaves when the link is noisy rather than gone.
That may be true, but it seems to strengthen the case for moving the human out of the forward cockpit rather than keeping them there.
If the unmanned aircraft are the ones flying far ahead, taking the risk, and extending the standoff envelope, then why is the human still sitting in the forward fighter rather than supervising from a safer node further back?
At that point it seems like the architecture is optimizing for tactical latency and current doctrine, not necessarily for the cleanest end-state.
At 10-mile intervals you're maintaining a high-bandwidth, low-latency mesh network in a contested electronic environment. If the command aircraft is 10 miles away and the enemy is jamming the link, the drone is going to be making split-second (potentially) lethal decisions without the pilot.
You're right about them both costing about the same, so the real leverage only comes if these drones can stay out of the engagement envelope while sending cheaper submunitions (likely using something like these Ragnaroks (~$150k) https://www.kratosdefense.com/newsroom/kratos-unveils-revolu...) to do the actual baiting.
This seems to be the generally agreed upon direction defense companies are going, but a couple architectural concerns come to mind regarding this "Manned-Unmanned-Teaming" approach:
- Even if the XQ-58 has a low radar cross section, a swarm of four drones flying in formation with a non-stealthy Eurofighter significantly increases the aggregate probability of detection. Unless these drones are performing active electronic countermeasures or "blinking" to spoof radar returns, they’re essentially a giant "here we are" sign for any modern radar. I wonder if they've compensated via the flight software to manage formation geometry to minimize the group's total observable signature?
- Anti-air systems will prioritize the command aircraft (the Eurofighter) immediately. If the C2 link is severed (kinetic kill, high-power jamming) what is the state-machine logic for the subordinates? Do they revert to a fail-passive (return to base) or -active (continue last assigned strike) mode? Without a human-in-the-loop, rules of engagement issues are abound. (I'm not even accounting for the fact that the drones probably rely on calculations from the command craft, so edge-computing will factor in as well.)
- They're calling these "attritable," but at $4M a pop plus the cost of the sensors, they aren't exactly disposable. Is the cost-per-kill for an adversary’s interceptor missile actually higher than the cost of the drone it's hitting?
They don't call the base model 'iPad Neo.' They just call it iPad (https://www.apple.com/ipad-11/). It's the same market segment and even uses the exact same color palette.
They also just established that 'e' is the designator for budget model (https://www.apple.com/iphone-17e/) so best guess is they thought 'MacBook e' looked strange so instead it's 'nEo'. And don't forget the 2004 eMac.
Looks like they're using some new variant of branding font for this. Inspect Element shows it as SF Pro Display, but it's actually just being masked over with an image
My first thought was, "So, Replit and ilk?", seems they expected that comparison:
> How is Glaze different from Lovable, Replit, or v0?
> Those tools build for the browser. Glaze builds for your desktop. That means your apps can access your file system, your camera, keyboard shortcuts, menu bar integration, and background processes. Things a web app can’t do. It’s a different category entirely.
Pretty sure modern web apps can do all of those (sans menu bar). (If anything they do background processes better since you can send a very long task off to a server and shut off your computer, come back later and pick up where you left off.)
Also, as others mentioned, this just seems like Claude Code with extra steps, unless they managed to nail some sort of design standard enforcement they feel is better than what most people can get out of it.
The quick publishing is kind of nice, but it immediately made me think it would be more interesting to have a way to quickly remix other people's creations, similar to the Figma Community tab: you can take someone else's work, break it apart to see how it works, then tweak it how you want it.
I think this is exactly the crux: there are two different UX targets that get conflated.
In operator/supervisor mode (interactive CLI), you need high-signal observability while it’s running so you can abort or re-scope when it’s reading the wrong area or compounding assumptions. In batch/autonomous mode (headless / “run overnight”), you don’t need a live scrollback feed, but you still need a complete trace for audit/debug after the fact.
Collapsing file paths into counters is a batch optimization leaking into operator mode. The fix isn’t “verbose vs not” so much as separating channels: keep a small status line/spine (phase, current target, last tool call), keep an event-level trace (file paths / commands / searches) that’s persisted and greppable, and keep a truly-verbose mode for people who want every hook/subagent detail.
I'll look into what's going on with some of the other browsers.
To clarify, the game actually runs a quick validation when the timer runs out to check if your word is valid. If it is, the ball returns automatically—so you don't have to hit Enter or Space, but doing so early gives you a speed bonus.
As for getting rid of Enter/Space entirely, auto-submitting can be tricky with compound words (e.g., should it submit 'REGULAR' or wait for 'REGULARLY'?).
I was suspicious when I first coded this that it needed a better way to introduce the rules. I've gone ahead and buffed the HINT so that time doesn't resume until you get a chance to type out your first word. I also added a background hint on the return phase instead of just leaving it blank.
Another comment mentioned that the instructions that I put on itch.io made it clearer (and forgot to post on HN, whoops!) so I've pasted them below.
-------------------
HOW TO PLAY
1. Type HIT to serve
2. Type your opponent's word to line up your character with the ball, then type your word to send a volley back
3. Submit your word before the time runs out. The faster you submit your word, the faster your hit!
I recently rewatched a Tested Q&A where Adam Savage discussed his post-Mythbusters life; his framing of that duality was very similar: https://youtu.be/2tZ0EGJIgD8?t=322.
It aligns with a common design principle: constraints often make a problem space easier to navigate. I suspect life is similar. Having limited time creates a "specialness" that is easily lost when you suddenly have an infinite amount of time at your disposal.
https://thomas.design
https://triglavis.com
https://tryworkspaceos.com