Indie dev in Toronto shipping AI products in public. Building Slideforge, Otraasi, FixMyLaw and OnTheTower. Honest notes on what AI agents can and can't do yet.
Tesla just revealed the Roadster. Nine years late, and it came with rocket thrusters.
At an event outside Waco, Texas, Musk showed a SpaceX-engineered Roadster driving up an inverted ramp, spinning mid-air, and easing itself down onto the pavement. It also lifted itself over a Cybercab, Cybertruck, Semi, Model Y and Model 3 while moving.
What's confirmed:
โข 0-60 mph: Tesla showed it in 1 second
โข Lift system: 10 nozzles fed by 600 liters of compressed air, and the tank can be recharged (per Lars Moravy, Tesla engineering VP)
โข Production: Musk says Tesla will build it in 2028
โข Price: not announced
โข Reservations: $50,000 deposits, and some 2017 buyers paid $250,000 for a Founders Series spot
What's still missing: range, battery size, top speed, and a final price. The 2017 promises were 620 miles, 200 kWh, 250+ mph and $200K, and none of those have been reconfirmed.
Musk also floated trying the Snake River Canyon jump that Evel Knievel attempted.
My take: this is a halo car and a SpaceX x Tesla flex, not a spec-sheet launch. Musk himself admits it may not move revenue much. The demo was wild, but until there's a price and a date I can trust, the reservation holders are still waiting.
#Tesla #SpaceX
Source: https://t.co/z8yiGIJWWQ
Patch alert: another Cisco Secure Email Gateway zero-day is under active attack (CVE-2026-76519, CVSS 9.8).
One crafted attachment can give an attacker root if File Reputation is on. No workaround, so update now and check logs off the box.
Source: https://t.co/wSLRZo7LVm
United now has Starlink Wi-Fi on 47% of its daily flights, with 639 planes equipped.
Its new tracker tells you if your flight has Starlink when you enter the flight number up to 3 days out. Whole fleet by end of 2027. #Starlink
Source: https://t.co/Mydq7HKwhZ
Barclays says a SpaceX-Tesla merger could be announced before Tesla's annual meeting (due by Dec 6), likely SpaceX buying Tesla in an all-stock deal.
Its math: Musk would own ~37% of the combined company but hold ~56% of the votes. #Tesla
Source: https://t.co/2Hmo0QMcG7
Apple is inviting media to an all-day "launch experience" in LA on Oct 28, per Bloomberg's Mark Gurman.
Expected: the first touchscreen MacBook Pros (OLED, Dynamic Island), an OLED iPad mini and M6 Macs. No livestream expected. #Apple
Source: https://t.co/M6wcZhHrek
Withings now charges to sync your health data.
New accounts since Oct 13 need Withings+ ($99.50/yr) to auto-sync to Apple Health and Health Connect. Older accounts keep it free, even on new devices.
Moving your own data now costs extra.
Source: https://t.co/6S4ChEK07s
Musk on Ukraine's Starlink slowdown: Starshield, SpaceX's military network, is owned and controlled by the US government, so its speed limits are set by the US military and ultimately the president.
Starlink stays civilian and defensive only.
Source: https://t.co/Gh1YEpoPOT
CoreWeave's 5-year credit default swaps hit a record ~1,194 bps today, pricing in a 60%+ chance of default within five years.
Jensen Huang says CoreWeave's "strength is improving every day" and Nvidia doesn't need to backstop it.
$CRWV
Source: https://t.co/vV8Ip4ue19
Two workers have died after an explosion at Intel's Kiryat Gat chip fab in Israel. A third is in serious condition.
Early findings point to a rare failure on a tank holding ~8 tons of nitrogen trifluoride, a toxic gas used in chipmaking.
Source: https://t.co/fsFZisgguX
Bluesky's AI app Attie now does "social vibe coding": anyone can build sites, games, quizzes and polls for their community on ATProto.
Bring your own AI model or use open-weight ones, then share to Bluesky's 47M users.
#vibecoding
Source: https://t.co/HxZxITsWey
@talirezun Agents running overnight eat quota fast. Cap their loops, give each a narrow scope, and restart sessions often so context doesn't snowball. Routing routine steps to a cheaper model saves a lot.
@solomonuche Weekly caps hurt. Try planning in a cheaper model, then let the stronger one only execute. Also clear context between unrelated tasks and point it at specific files, since huge context is what burns limits fastest.
@khakha_x@bcherny@thsottiaux Weekly limits drain fast with long contexts. Start fresh sessions per task, use /compact often, and keep a short CLAUDE.md so the agent doesn't re-read the whole repo each time. Smaller models for simple edits stretch the quota.
NVIDIA open-weights Nemotron 3.5 Super VL today: a 120B multimodal MoE (12B active) for document/chart/agent work, plus Bolt 74B-A9B for local agents in 24GB. Partners say post-training lifts specialized tasks past Claude/GPT baselines.
Source: https://t.co/lIxm1ASMEt
Simons Foundation, Anthropic, and OpenAI are putting up a combined $15M so mathematicians can vet and build on AI-generated proofs - workshops, tools, and compute, with the foundation running the open call. OpenAI's share: $5M.
Source: https://t.co/DKbyNKfQ7l