Key Takeaways
- Perplexity released a Decisions API powered by a multimodal decision model that outputs probability distributions over fixed answer sets, priced at $0.04 per million input tokens with an 85.71% benchmark score.
- SpaceX completed Crew-13 launch and ISS docking, a Falcon 9 Transporter-18 mission with 130 payloads, and a Falcon Heavy NROL-97 launch, totaling four booster landings.
- Basecamp launched Bonfire and opened agent collaboration across todos, cards, files, and messages, with agents replying in conversation threads like humans.
- YouTube will adjust Shorts recommendation algorithms to prioritize original content and reduce distribution for re-upload-heavy channels.
- Shopify launched Canvas for store design, with the team emphasizing Liquid remains the underlying driver.
- Grok Bot received updates including proactive suggestions, faster responses, image and text recognition, and coding task handoff to Cursor.
- Perplexity plans to open-source its multimodal decision model and make Decisions API output tokens free, with further price cuts planned.
1. AI Models and Infrastructure
- Perplexity launched the Decisions API, driven by the multimodal decision model pplx-decider-v1-27b, which outputs probability distributions over fixed answer sets rather than text. The API is priced at $0.04 per million input tokens and scored 85.71% on benchmarks. The author stated the model will be open-sourced, Decisions API output tokens will be free, and further price reductions are planned in coming days. Perplexity also released a Jev alternative on Hugging Face and solicited feedback on improving its developer platform. — via 1 2 3 4
- Grok Bot received updates enabling it to proactively suggest help, respond faster, recognize images and text in chats, and respond when directly mentioned. It can take over coding tasks and hand off to Cursor, manage PRs through GitHub and Origin plugins, and share build video demos. SpaceXAI published its engineering team's internal Bot templates covering engineering leads, QA, overnight code audits, project management, research fact-checking, supervising cloud coding agents, and code writing and issue triage. — via 1 2 3 4 5 6
- A viewpoint stated that if tasks are verifiable, AI has now solved them: 18 months ago the best models scored about 37% below average accountants, while today models can achieve perfect scores. Opus 5.5 reportedly solves all accounting tasks almost instantly with 100% accuracy, while humans take more time and make more errors. — via 1 2
2. Space and Aerospace
- SpaceX completed the Crew-13 crewed launch and docked with the International Space Station, while Falcon 9 executed the Transporter-18 mission delivering 130 payloads to orbit, and Falcon Heavy completed the NROL-97 launch, NRO's first use of Falcon Heavy, with four booster landings total. — via 1 2 3 4 5 6 7 8 9 10 11 12 13
- Starlink in-flight Wi-Fi progress: Alaska Air Group has installed it on over 40% of its fleet, Hawaiian Airlines' long-haul fleet and Alaska Airlines' regional fleet are fully completed, and one tweet stated in-flight Wi-Fi will launch in 2027. — via 1 2
- SpaceX's Star Mind superintelligent spacecraft will be equipped with over 250kW of solar power, about 20% higher than the International Space Station. A cited tweet stated satellite deployment sizes roughly double each generation: Starlink V1.5 at about 11 meters, V2 Mini at about 30 meters, V3 at about 60 meters, and AI1 wingspan at about 75 meters. — via 1
3. Developer Tools and Platforms
- Basecamp launched Bonfire, a feature explored over a decade ago but never released, originally designed for the Campfire chat tool. Campfire was merged into Basecamp and Bonfire now belongs to it, available to all Basecamp 5 customers. The author stated that 22 years after launching Basecamp, it has become the best framework for humans and agents to collaborate in one place: @bitsweat assigns tasks to agent Clawdito in Basecamp, gets replies, requests follow-ups, and creates PRs; @jasonzimdars collaborates with agent Marie and his own agent JZ bot in message board comment threads. This capability applies anywhere in Basecamp where people can communicate asynchronously or in real time, including todos, cards, files, documents, and photos, with agents replying in conversation flows like humans and retaining thread context. The team has been running and refining it in their own accounts, and anyone can use it via CLI, with official plans to simplify creating agent users within Basecamp. — via 1 2 3 4
- Guillermo Rauch proposed "verification engineering" as the future direction, covering proofs, end-to-end testing, benchmarks, and linters, and believes testing will split into deterministic and agentic categories. He shared e2e, an agentic testing framework for any application, initialized via
npx e2e init, fully open source, supporting mixed deterministic and agentic APIs, covering web and mobile, with bring-your-own agent and infrastructure, and local or CI execution. He also said SvelteKit 3 performs excellently and he looks forward to Async Svelte and Remote Functions; his small app built and deployed end-to-end in just 15 seconds, with fx and opus easily completing all the work. — via 1 2 - YouTube will adjust its short-video recommendation algorithm to further prioritize original content and reduce distribution of re-uploaded content without added value, meaning channels primarily focused on re-uploads will see reduced distribution in Shorts feeds. A cited tweet stated YouTube aims to highlight content with real value and discourage "minor technical edits" and "templated" changes, which basically covers the industry's clip-reposting practices. The tweet commented that this will offend some who built careers on others' content, mainly clip accounts, but the change itself is good and will address long-standing overall quality issues. — via 1
4. Product Launches and Business Updates
- Shopify launched Canvas, which the team said exceeded expectations, making online store design fun and democratized, while emphasizing it is still powered by Liquid. A cited tweet said that 12 years ago when building the first store for @kotn, even with programming skills it took two weeks to make a decent version and years to be satisfied, and Canvas is the tool they wished they had then. Shopify is said to "over-invest" in projects like Editions to make the web more beautiful, and Canvas continues that goal: given AI easily generates homogeneous content, the team spends significant time guiding Sidekick to make good design decisions while giving users space to create unique work. — via 1 2
- Shopify ranked 3rd on Forbes' World's Best Employers list, with a cited tweet calling it "the company at the forefront of AI" where employees love working. Shopify Racing's No. 11 LMP2 car will race this weekend at Road Atlanta's Petit Le Mans for its final race of the year, with @tobi and @MathiasBeche co-driving, targeting a podium finish. — via 1 2
- Strike reduced trading fees globally, now as low as 0.89%; large orders can access competitive execution prices and dedicated support through Strike Private via [email protected]. — via 1
- Perplexity released a Jev alternative on Hugging Face, and the author said he wants to hear how to improve its developer platform; a cited tweet invited users who tried Perplexity's various APIs (Search, Agent, Embeddings, Decision, etc.) and were disappointed to provide feedback via DM. — via 1 2
