Grok 4.7 Lands on Amazon Bedrock and Gemini Enterprise as Starlink Expands Across United Airlines

Key Takeaways

1. Grok 4.7 is now available on Amazon Bedrock and Google Gemini Enterprise Agent Platform, with a 500K context window and $2/$6 per million input/output tokens on the xAI API.
2. Starlink is set to equip 880+ United Airlines aircraft by end of 2026, with full-fleet connectivity expected by end of 2027; it has already connected 14.2 million devices across 464,000+ flights.
3. SpaceX holds the top five fastest ISS docking times, all using Dragon spacecraft, led by Crew-13 at 7 hours 55 minutes.
4. Tesla took the top two spots in the Dutch new-car market in September, with Model Y at 2,078 registrations and Model 3 at 1,087, totaling 3,165 units or 8.7% share.
5. Vercel has been migrating to Rust, starting with Turborepo from Go, but internal ROI debates persist; the company also released gdp-ts for compile-time API safety proofs.
6. Garry Tan highlighted that AI infrastructure primitives are commoditizing, while startup harnesses like Grep convert token usage into deterministic tested code, saving clients 250,000+ hours.
7. Levelsio built an AI story generator over a weekend but chose not to ship it, citing no moat, high inference costs (~$5 per 60-second episode), and a saturated market dominated by course sellers.

1. AI Models and Infrastructure

1. Grok 4.7 has launched on Amazon Bedrock and Google Gemini Enterprise Agent Platform, both of which previously offered Grok 4.6. The model targets coding and knowledge work, is described in documentation as SpaceXAI's strongest model to date, and offers a 500K context window on the xAI API with pricing at $2 per million input tokens under 200K prompt tokens and $6 per million output tokens. A separate repost noted that Grok Bot passed Google's "I'm not a robot" test at launch, while other similar agents have not yet shown video proof of doing so. — via 1 2
2. OpenRouter co-founder Alex Atallah argued that the "agent loop + notifications, third-party connectors, context management, memory, sandbox, agentic web search, persistent agents" stack being built across the industry is just new table stakes, similar to how everyone built databases, user tables, and login pages in 2005. He believes AI also has foundational needs and that differentiation space remains large. Garry Tan added that when everyone builds the same primitives, those needs only intensify and will eventually converge on the correct OS. — via 1
3. Garry Tan argued that lab-produced harnesses have an incentive to burn tokens, making startup harnesses practically useful. He cited Grep as an example that observes agent usage to automatically convert token consumption into deterministic, repeatable, tested code. A quoted post described Grep building self-improving agents for clients like IG to handle compliance alerts and other repetitive operations, with over 80% of IG traders completing account opening without human intervention; its AgentRun harness lets repeated steps settle into code and workflows, reserving frontier intelligence for exceptions, with clients saving over 250,000 hours of operational work cumulatively. — via 1 2

2. SpaceX, Starlink, and Tesla

1. Starlink will be installed on more than 880 United Airlines aircraft by the end of 2026, with full-fleet connectivity expected by the end of 2027, according to a repost by Elon Musk. Starlink has already connected 14.2 million devices across more than 464,000 flights and is currently active on over 560 United aircraft. Musk also cited a passenger case of streaming five college football games simultaneously on a commercial flight using Starlink across three seatback screens and two laptops, noting United already has over 500 aircraft equipped. — via 1 2
2. SpaceX swept the top five fastest US launches to ISS docking, all using Dragon spacecraft: Crew-13 at 7 hours 55 minutes (October 1, 2026), CRS-31 at 12 hours 23 minutes, Crew-11 at 14 hours 43 minutes, Ax-2 at 15 hours 35 minutes, and Crew-4 at 15 hours 44 minutes. — via 1
3. Tesla took the top two spots in the Dutch new-car market in September, with Model Y registering 2,078 units for first place and Model 3 registering 1,087 for second. Tesla totaled 3,165 registrations that month, representing 8.7% of the overall Dutch new-car market. A separate repost argued that Tesla's practice of iterating and scaling its own inference computer for every car sold will look unprecedented in hindsight, and that the current compute shortage is only the tip of the iceberg, with the real shortage arriving when autonomy becomes indispensable. — via 1 2

3. Developer Tools and Startup Strategy

1. Guillermo Rauch disclosed that Vercel has been advancing Rust adoption for some time, with Turborepo among the earliest projects migrated from Go to Rust. While the migration completed, internal ROI debates remain significant: Rust is better suited for low-level OS access and critical for build systems, but human migration costs are high; Go is fast, well-designed, and easy to iterate, and the team struggled with the fact that humans write code even knowing Rust was the better choice. Rauch also said the metric has shifted so that "best for humans" no longer necessarily equals "best for business," and that Rust may not be the ultimate toolchain since it was designed before the agent wave. — via 1
2. Rauch released gdp-ts (Ghosts of Departed Proofs for TypeScript), including a library, linter, and AI skill for safer API design: sensitive functions require callers to provide proof that authorization checks were performed, verified by the type checker at compile time to prevent teams and agents from introducing severe security bugs. He noted such patterns have long existed in Haskell and other ecosystems but remained niche due to manual code review, cognitive, and syntax overhead; now that agents write code beyond human review capacity and work better in tight loops with hard constraints, the situation has reversed. The README and examples model Vercel's real API constraints, such as requiring proof of specific role and entitlement to change a Project password; research comes from Matt Noonan and Ollie Charles. — via 1
3. Levelsio built an AI story generator over a weekend that takes a prompt and generates stories with conflict, suspense, and narrative arcs, splitting them into episodes, storyboards, characters, music, posters, voiceovers, and visuals, with multiple styles and slideshow or video rendering. He ultimately decided not to ship it, concluding that such "AI glue" apps have extremely low build barriers and no moat, with the real barrier in GTM strategy; the space is extremely saturated with many AI film tools using the same models differentiated only by thin prompt or UX layers; one of the largest advertisers in the space sells "how to make AI films" courses, suggesting hype and course sales are more profitable than the product; costs are high at roughly $5 in inference alone for a 60-second mini-episode, margin would be eaten by competitors, and user refunds or chargebacks could bankrupt a company; he felt he was just producing "AI slop" and did not want to participate. — via 1
4. Garry Tan reposted that using Opus 5.5 over a weekend, he ported the Bungie game Myth from his high school days to WebAssembly, theoretically supporting the full single-player campaign and PvP multiplayer. He said the ease of porting surprised him, leading him to realize that "abandonware" no longer exists and any program can run on any hardware at very low cost. — via 1 2

4. Market Signals and Observations

1. Levelsio noted that Airbnb's rating range has narrowed to 4.5 to 5.0. He analyzed booking platform rating distributions: Airbnb spans only 4.5 to 5 with a median of 5; most platforms range from 4 to 5 with a median around 4.25, and few scores fall outside that range; he considers Yelp's ratings the most honest, showing a true normal distribution; his own Hotelist has a median of 3.5, matching the goal that "most stays are mediocre." He clarified he is not lobbied by the hotel industry and was once an Airbnb loyalist preferring kitchen-equipped rentals for cooking, but untrustworthy ratings caused "rating anxiety" and too much variance in stay quality, making Airbnb unreliable outside Asia and pushing him back to hotels; he believes hotels could capture significant Airbnb market share by adding kitchens to rooms. — via 1 2 3
2. A repost said that inspired by Levelsio's post on Airbnb rating distortion, someone developed a Safari extension that stretches the 4.6-5.0 rating cluster to a 0-10 range for comparison, saying "4.9 is just the new 3," published on GitHub and coming soon to the App Store. Levelsio also said Airbnb's PR team is "scrambling," calling the response typical PR talk. — via 1 2
3. Yohei reposted that emerging fund managers back "outlier" projects at more than twice the rate of large funds on a single investment; combining five emerging managers' portfolios significantly broadens investment scope, while combining more homogeneous large funds does not. The quoted view argues early-stage VC relies on discovery, large funds prefer dominating existing categories rather than driving discovery, so the market will continue to depend on small and emerging managers for discovery. — via 1
4. Tobi Lutke noted Canadian projects under construction: Churchill Falls at roughly the equivalent of 9 nuclear reactors in clean power; the Pacific Link oil export route at a scale close to UK national consumption; and LNG Canada Phase 2 gas volumes at about half of Canadian domestic consumption. He also said he hears many similar stories of consumers who used to buy on Amazon now buying directly from brands through Muse, which he sees as positive for Shopify. — via 1 2