NZ ☄️

AI Insider / Tracking the race to AGI
Startups • Agents • Open Source
Sharing what matters before it goes mainstream

[NZ ☄️ @CodeByNZ]

@Tesla unveiled the new Roadster less than a day ago, so I built a game where your Tesla can fly 🚀

▎ Hop over traffic, flip in mid-air, grab nitro and dodge honking drivers

Play free in your browser 👇
▎ https://t.co/lszLXsE0eG

How it works

▎ 🚗 Flying Tesla is a browser game inspired by the Roadster unveil.

▎ Pick a driver: Elon, Jeff, Sundar, Jensen or Steve, each with a perk.

▎ Fly: tap FLY or press Space to hop over cars. You can't fly again right away, so time your jumps.
▎ Flip: press ← → mid-air for front flips and backflips. Each flip you land is +300.
▎ 3 lanes: switch lanes with ↑ ↓ or a swipe.
▎ Nitro: grab blue N₂O canisters to boost and smash through traffic.
▎ Gifts: collect boxes for Grok and Cursor tokens. Some float, so you have to fly to get them.
▎ 8 camera views, including first-person from the driver's seat.
▎ Easy, Normal and Hard modes, with levels that speed up.

Works on desktop and mobile. Fan-made, not affiliated with @Tesla , @spacexai

[Elon Musk @elonmusk]

Turns out the future does have flying cars

[NZ ☄️ @CodeByNZ]

@rubenssoto_ai that's fair, and honestly for most workflows there probably isn't one. the advantage only shows up when tasks can run in parallel or the context gets too big for one window. otherwise you're paying for handoffs you don't need

[NZ ☄️ @CodeByNZ]

Anthropic just told developers to stop stacking subagents and fold them into one agent with skills.

And honestly, I think the timing says a lot.

A year ago, Anthropic's own research setup was a lead agent directing subagents, at roughly 15x the tokens of a normal chat. Now they say one agent is often faster and cheaper, and they cut Claude Code's system prompt by over 80%.

Same month, the S-1 draft reportedly says Anthropic is running into compute limits, and power users keep reporting that Opus 5.5 limits drain faster than ever.

Maybe this is purely about building better agents. But it also looks like a company teaching everyone to use fewer tokens.

[NZ ☄️ @CodeByNZ]

@johnroodepic the rebuild cost is a good point. each subagent starts cold and burns tokens working out what the lead already knew. shared notes help a lot, but the handoff still costs something, since every session has to read them before it does anything

[NZ ☄️ @CodeByNZ]

@bluvi_hq i'd say both can be true. the advice is probably right for most setups, it just also happens to save them a lot of compute at the exact moment they need it

[NZ ☄️ @CodeByNZ]

@srcbuild kind of, but the difference matters. a skill is just instructions loaded into the same agent when it needs them, a subagent is a whole separate run with its own context. so skills are cheaper but they fill your context, subagents protect it but cost more

[NZ ☄️ @CodeByNZ]

@Compilethings makes sense for long runs. you're paying more tokens overall but buying a main session that doesn't fill up and drift. curious if you've noticed the orchestrator missing details from what the subagents send back

[NZ ☄️ @CodeByNZ]

@rubenssoto_ai cheaper per token doesn't mean cheaper per task. you pay for the handoffs, the repeated context and the coordination, and that's usually where the bill comes from

[NZ ☄️ @CodeByNZ]

@hakimieiqbal they didn't list the exact sections. what they said is the system prompt for newer models is over 80% shorter, and the logic is that skills now load only when needed instead of sitting in the prompt all the time

[NZ ☄️ @CodeByNZ]

@TokenGremlin @grok bro removed both 😹

[NZ ☄️ @CodeByNZ]

@blueemi99 @TokenGremlin what’s 6k ? let’s say 10k

[NZ ☄️ @CodeByNZ]

@Chris_Wozniczek @DevinAI devin was early on the idea, but the models in 2024 weren't reliable enough for a manager agent to trust its workers. the pattern only really works now

[NZ ☄️ @CodeByNZ]

@TokenGremlin people saying the agent side isn't reliable yet are missing that a scam doesn't need reliability. it only has to work on 1 in 1000 calls to pay

[NZ ☄️ @CodeByNZ]

@DanielGPT2022 half true, the models write a lot of the code now. but someone still has to decide what gets built and keep it safe, that part isn't automated yet

[NZ ☄️ @CodeByNZ]

hot take: what if these Anthropic and OpenAI employees are leaving to join SpaceXAI to work on Grok Bot 👀

[NZ ☄️ @CodeByNZ]

@0xHvdes lol that would be something. honestly with the talent moving around right now, nobody's lead is safe by 2027

[NZ ☄️ @CodeByNZ]

@0xHvdes looks relooks real to me

[NZ ☄️ @CodeByNZ]

@rubenssoto_ai cheapseek looks real than the others for me

[NZ ☄️ @CodeByNZ]

ME AFTER 1,000 PEOPLE LONG PRESS LIKE ON MY POST FOR 6 MONTHS.

[Elon Musk @elonmusk]

We’re adding long press on like to send someone $1 via 𝕏 Money

[NZ ☄️ @CodeByNZ]

@JamesMalsawm Alexandr is doing everything to push Muse

[NZ ☄️ @CodeByNZ]

Tibo got MOGGED by Alexandr Wang...

never a dull moment on this app, they weren't lying when they called it the everything app 😂

[NZ ☄️ @CodeByNZ]

@aashatwt it’s a lot fr

[NZ ☄️ @CodeByNZ]

@devdiary0x fr man, we just watch