Copilot moves into JetBrains Air, AI agents move into game engines, and Anthropic shows a $17B quarter

Nov 3, 2026

AGENTS: GitHub Copilot now runs inside JetBrains Air, locally or in cloud sandboxes

GitHub announced on Nov 3 that Copilot is now an agent you can pick inside JetBrains Air, connected through the open Agent Client Protocol (ACP). You delegate a coding task, steer it while it runs, and review the changes next to your code, with several agent sessions tracked in parallel. It uses your existing Copilot subscription, not JetBrains AI credits. Cloud tasks run through Air Teams: each one gets a Docker container on a fresh virtual machine, all traffic goes through a proxy, and no state carries over between tasks, so Copilot can build, test and install packages without touching your machine. There are limits worth knowing: you cannot choose per task whether your personal or organization Copilot plan pays, Air's analytics do not show Copilot usage, and Air admins can switch Copilot off but cannot set Copilot spending limits. Cloud tasks need an Air Teams business organization. For AI coders, it is another sign that ACP is turning agents into plug-ins you can swap between editors.

Source: https://github.blog/changelog/2026-11-03-github-copilot-is-now-available-in-jetbrains-air/

Oct 31, 2026

AGENTS: DeepSeek open-sources a prototype "super-agent" that picks up coding tickets from GitHub chats

DeepSeek's Harness team published dsh-superagent-prototype on GitHub under the MIT license, released Oct 31 according to TechTimes' Nov 3 report. It watches one GitHub repository and, optionally, Feishu chats. When an authorized person asks for help, it opens a work ticket, starts a separate agent session in its own working directory, and replies in the original thread, often with a pull request. A web panel lists tasks that need a human. It only acts when someone asks or a team rule allows it. DeepSeek is blunt about its status: the repository calls it a research prototype "for reference only, not an official DeepSeek product", says it may contain bugs, and says the public code differs from its internal version, so some designs in the accompanying research article are not implemented. Setup needs a dedicated agent GitHub account and an authenticated gh CLI. Treat it as a blueprint for building your own always-on coding agent, not a drop-in Claude Code replacement.

Source: https://github.com/deepseek-ai/dsh-superagent-prototype

Nov 2, 2026

GAME DEV: Kimi launches a game-dev solution whose agents work inside Unity, Unreal, Godot and Blender

Moonshot AI's Kimi introduced its game development and creative work solution, announced Nov 2 per IT Home. It combines the Kimi K3.1 models, the Kimi Code and Kimi Work agents, and plugins that let the agents edit real projects in Unity, Unreal Engine and Blender. Kimi says designers with little programming experience can get a playable Godot prototype in one or two days, and that the agent runs the build, reviews renders and gameplay recordings, and fixes issues before handing it over. On Unity it works with Unity China's Tuanjie Codely Agent, and Kimi says one demo ran autonomously for 20 hours to recreate a game as a native engine project. Its showcase scenes include a cathedral interior of about 3.3 million triangles in Blender and an Unreal Engine 5.8 palace with 36 million source triangles. These are Kimi's own demos, not independent tests. It also covers LiveOps work such as ad-creative variants, localization drafts and player feedback triage. For indie game devs, the useful idea is its advice to start with one small, measurable task, like prototyping a single mechanic.

Source: https://www.kimi.ai/news/kimi-game-solutions

Nov 3, 2026

GAME DEV: Xogot brings an AI coding agent to Godot on iPhone, iPad and Mac

Xogot added a built-in AI agent on Nov 3, written up by Miguel de Icaza. Xogot is a native version of the open-source Godot engine for Apple devices, and the agent can now write and modify code, debug problems and make changes directly in your Godot projects on iPhone and iPad, with more features on the Mac. Xogot does not supply its own AI server: you add a provider in AI Settings, using an existing subscription or API key for Claude, ChatGPT or GitHub Copilot, a key from another provider, or a local model you host yourself. The example prompts are the everyday jobs of a solo developer, such as "Why isn't this enemy detecting the player?", "Add touch controls to this character" and "Help me reorganize this inventory system." The team says it also used the agent to fix CI builds and merge large conflicting branches. It is available now on the App Store. For indie game devs, it means you can clear small Godot tasks from a tablet, with whichever model you already pay for.

Source: https://blog.xogot.com/a-helping-hand-for-game-creators/

Nov 2, 2026

BUSINESS: Anthropic's amended IPO filing shows $17B in Q3 revenue and $895B in compute commitments

Anthropic filed Amendment No. 1 to its Form S-1 on Nov 2, adding third-quarter numbers, as CNBC and Reuters reported. Revenue was $16.96 billion for the quarter, up about 1,127% from a year earlier and 47% from Q2, although growth slowed from 144% the quarter before. Revenue for the nine months to Sept. 30 was $33.24 billion. Anthropic swung to a $567 million operating loss from $974 million of operating income in Q2; per StockAnalysis' breakdown, the filing blames $5.0 billion of compute bought on short-term commitments, which costs significantly more than its multi-year contracts, plus $2.0 billion of non-cash charges. The $23.29 billion net loss came mostly from non-cash changes in the value of convertible notes and derivatives. It has about $895.4 billion in purchase and lease commitments, substantially all for compute, and warned that some providers depend on outside financing, so contracted capacity could be delayed or repriced. The price range and share count are still blank; the planned ticker is ANT on Nasdaq. For AI coders, this explains why usage limits keep tightening: compute is the cost that decides your plan.

Source: https://www.cnbc.com/2026/11/02/anthropic-ant-s-1-quarter-financials.html

Nov 3, 2026

DEALS: fal doubles its valuation to $16B with a $1.3B round; Traversal's AI site reliability agent becomes a $1B unicorn

fal announced on Nov 3 a $1.3 billion Series F led by Bessemer Venture Partners, with NVIDIA, Sound Ventures, HOF Capital and T. Rowe Price, at a $16 billion valuation, double its $8 billion mark in April. It also disclosed that April's $350 million Series E was led by Sequoia, for $1.65 billion across both rounds (TechCrunch). fal runs image, video and audio models behind one API, and game developers are among its production customers. Bloomberg reported it passed $800 million in annualized revenue earlier this year, and CEO Burkay Gur said demand got so high in the summer that it had to rate-limit customers. The money goes to compute and hiring beyond its 185 staff. Also on Nov 3, Traversal raised $120 million at a $1 billion valuation, co-led by Kleiner Perkins, Redpoint Ventures and Khosla Ventures, with Sequoia, Battery Ventures and Amex Ventures. Its AI agents investigate and help fix production outages for customers including Nubank, Harvey, DigitalOcean and PepsiCo, and it says ARR grew 9x this year (company-reported). The signal: the companies serving and protecting AI-written code are where investors are writing the biggest checks.

Source: https://fal.ai/blog/series-e-f

Ship faster this week

Grab the free Agentdex starter kit with prompts and agent configs for Cursor, Claude Code and Codex.