In this edition, covering 10 to 16 November:
- Google launched Gemini 4 Hydrogen, a faster model for coding and agentic work priced at $1.50 per million input tokens and $7.50 per million output tokens, and opened Gemini 4 Carbon to developers through paid APIs.
- Mirendil raised $1 billion in a Series A led by Kleiner Perkins and Clearlake Capital, and CodeRabbit says it passed $100 million in annual recurring revenue.
- AWS released AWS Emulator, a free tool that simulates more than 200 AWS services on developers' machines and that AWS says autonomous AI agents built and maintain.

This spot is open for 2027. One company puts its name on every AI Programming Weekly, which has 3,800+ subscribers. See what's included
Google
Google introduced Gemini 4 Hydrogen, a faster model for coding and agentic workflows with a 1 million token context window, priced at $1.50 per million input tokens and $7.50 per million output tokens. It is rolling out across the Gemini API, AI Studio, Android Studio, Google Antigravity and other Google products, and a new Orchestrator in Antigravity switches between Hydrogen and Gemini 4 Carbon during a task. Gemini 4 Carbon is now available to developers through the paid API tier in Google AI Studio and Android Studio, with pricing starting at $2 per million input tokens and $10 per million output tokens through 30 November. Gemini 4 Hydrogen · Gemini 4 Carbon
Google DeepMind researchers used AlphaProof Code to formally verify key security properties of Signal's cryptographic protocol implementation in four weeks. The system generated more than six million lines of Lean proofs and identified known bugs in older code, and Google DeepMind is releasing the verification artifacts. AlphaProof Code
OpenAI
Holiday Extras used OpenAI Codex to rebuild its main mobile app in two weeks, compared with more than six months previously. Gondola says its revenue more than doubled year over year after it adopted Codex and ChatGPT, while monthly engineering output rose 25% with no increase in engineering headcount. Holiday Extras · Gondola
Microsoft
Visual Studio Code 1.146 makes the Copilot harness the main agent experience and plans to remove the Local harness in February 2027. The release also adds an experimental Inbox mode, chat windows and a way to continue Copilot CLI sessions in VS Code. VS Code 1.146
Microsoft researchers report that poisoned fine tuning data and compromised model checkpoints can implant backdoors in coding agents. The backdoors activated through ordinary requests or repository content, reached up to 94.8 percent success on unseen repositories and could trigger unsafe tool actions such as command execution. Post training attacks
Funding and revenue
Mirendil raised $1 billion in a Series A led by Kleiner Perkins and Clearlake Capital for tools that use AI agents for frontier AI research, including its Mirendil-Base coding model. CodeRabbit says it grew from $10 million to $100 million in annual recurring revenue in just over a year, with more than 20,000 customers and over 3 million pull requests processed each week. Mirendil · CodeRabbit
Italian startup Nebuly raised $21 million in a Series A led by Planven to help large enterprises measure the cost and business value of AI agents. Erie Insurance's venture arm, Erie Strategic Ventures, invested in Blitzy, which develops AI software for modernising enterprise codebases, and in AI security company HiddenLayer. Nebuly · Erie Strategic Ventures
Coding agents and developer tools
AWS released AWS Emulator, a free tool that simulates more than 200 AWS services locally so developers and AI coding agents can test software without network access or cloud charges. AWS says autonomous AI agents built and maintain it, and lists known limitations including incomplete support for throttling, outages and regional differences. Expo launched Expo Simulators, cloud iOS simulators and Android emulators where agents such as Cursor, Claude Code and Codex reproduce mobile bugs and test fixes, at $0.10 per simulator minute on paid plans. AWS Emulator · Expo Simulators
Austrian developer Stanislaus Hees, 14, built Autark, an open source tool in which a manager agent assigns tasks to other AI agents; it has no sandboxing, approval prompts or permission gates. Y Combinator backed HyperBench launched agents that trace failures across software repositories and hardware components and test changes on real hardware. Autark · HyperBench
Katalon launched AI Test Runner, which runs existing natural language test cases in parallel on web and mobile apps, checks evidence for each step and returns videos and screenshots with every result; AI usage is included in the license cost. accessiBe turned accessFlow into an agentic accessibility suite for design, coding, quality assurance and production that will connect with Figma, GitHub, Jira, ClickUp, Asana and Linear, with people approving each AI remediation. Katalon · accessFlow
Bluehost launched AI Builder, which lets small businesses create websites and business applications through conversational AI without coding. Specialty insurer Fortegra is building applications for its managing general agent partners with insureMO's Atomic APIs and Agentic Engineering alongside its own AI assisted engineering teams. Bluehost AI Builder · Fortegra and insureMO
Surveys and benchmarks
Sonar's survey of 1,026 developers and engineering managers found that AI generates or significantly assists with 65% of committed code, while 76% say they cannot verify AI generated code as quickly as it is produced. TestMu AI, formerly LambdaTest, reports that 81% of surveyed teams merged AI generated code or tests they did not fully understand. In Uplevel's survey of 428 US engineering and technology leaders, 96% believe AI improves software development, while only 40% compare teams that use AI with teams that do not. Sonar · TestMu AI · Uplevel
New results from Toptal's ExpertCrowd-SWE benchmark put Anthropic's Claude Opus 5.5 first with 67.3%, while DeepSeek and SpaceXSI's Grok 4.6 scored 32.5% and 32.2%. Thoughtworks presented SpecBench, a platform that runs AI enabled software development workflows on defined tasks, evaluates the results independently and records time, token and cost use. ExpertCrowd-SWE results · SpecBench
Security
ENISA and CERT-EU used frontier AI models to review source code from 27 open source projects maintained by EU institutions and agencies and found 220 potential security issues; code owners confirmed 95% of the 88 findings assessed as true positives, and four CVE identifiers were assigned. Cycode launched Citizen Developer Security, which covers AI generated applications and unmanaged development environments of employees who build software outside engineering teams. ENISA and CERT-EU · Cycode
Agent platforms
At its MAX event in Miami, Adobe released a local Model Context Protocol server that gives ChatGPT, Claude and Gemini access to more Creative Cloud features, alongside new AI features including the Firefly Builder prototyping tool. Zendesk introduced Bridge, a headless platform that gives external AI agents controlled access to more than 400 tools through the Zendesk MCP server and connects with services including Codex, Claude Code and Lovable. Adobe · Zendesk Bridge
Charles Schwab introduced an MCP server that lets selected clients using Claude access account data, research and portfolio tools, with trading actions for eligible thinkorswim accounts. Square made Squaremates, Square MCP, Square CLI and Squaremates Apps available to US sellers at no additional cost. AppsFlyer expanded its Agent Hub with MCP tools that analyze AppsFlyer data and automate marketing actions. Charles Schwab · Square · AppsFlyer
Photon announced a ChatGPT plugin that lets licensed clinicians draft, review, authorize and send prescriptions using patient and treatment context. Luma Health introduced a ChatGPT plugin with 59 actions for healthcare workflows such as appointment preparation and rescheduling, which follows each health system's permissions and reads from and writes to the EHR. Photon · Luma Health
MCP connectors
Breach Secure Now launched a read only MCP connector for security program data in ChatGPT, Claude and Microsoft Copilot. Secure Passage added MCP server capabilities to Truman, its operating system for physical security, robotics and drone systems. IDrive launched RemotePC MCP, which lets ChatGPT, Claude, Cursor, Codex and Grok check computers, review screenshots and schedule restarts. Breach Secure Now · Secure Passage · RemotePC
LawVu says Claude, Microsoft Copilot, ChatGPT, Harvey and Legora can connect to LegalOS through its MCP server under each user's existing permissions. Sherweb's Cumulus MCP Server gives partners read only access to Partner Portal data, with write tools planned. RainFocus launched an Anthropic verified connector that gives Claude access to attendee, session and registration data for RainFocus powered events. Foxen launched the Foxen Connector, an MCP tool that lets multifamily property teams query renter risk data through AI tools such as Claude and ChatGPT, included for customers and broadly available in early 2027. LawVu · Sherweb · RainFocus · Foxen
Typeform launched an MCP server on all plans for creating forms and analyzing responses in ChatGPT and Claude. TrueDialog's MCP connector lets AI assistants send SMS and RCS messages, and SalesHood MCP brings approved sales content into AI tools with seller approval before content is shared. WellSaid Producer creates and edits voiceovers through MCP connected tools, and Clipto Studio lets AI assistants search indexed recordings of meetings and calls on Mac and Windows. Shopping app Shoplinkz launched the Ask Shoplinkz assistant and a Claude connector that lets Claude access users' Shoplinkz Mall Map and Lists and save products. Typeform · TrueDialog · SalesHood · WellSaid · Clipto · Shoplinkz
Cloud Native Computing Foundation
The CNCF added projects including OpenShell, Agent Substrate and llm-d, bringing its portfolio to 231 projects, and says the cloud native developer community reached 21.2 million. It also graduated NATS, an open source messaging platform used by more than 2,000 companies that connects applications, services and AI agents across cloud, on premises and edge. CNCF projects · NATS
Also
Altera launched Tera, an AI search and chat assistant powered by Vectara that gives FPGA engineers source linked answers from Altera's technical documentation. Tokyo based Licklider launched SciGround in public beta, a mathematical verification layer for LLM driven research that connects to MCP assistants. Altera Tera · SciGround