29 October 2026
Google launches Gemini 4 Carbon, its new top model, with a focus on cybersecurity
Google released Gemini 4 Carbon, and the press is split on how big the step is. HighLevel shipped several tools for building workflows and client apps with AI. Microsoft, OpenAI and Anthropic made moves aimed at big companies and at tracking AI costs.
Models
Google launches Gemini 4 Carbon with cybersecurity gains (opens the source in a new tab)
Google says Gemini 4 Carbon brings significant gains in coding, agentic tasks and cybersecurity. Google says it ranks first on the Artificial Analysis Intelligence Index (v5.0), ahead of Claude Opus 5.5 and GPT-6 Astra.
Why it matters: If you build on Gemini, plan a test of the new model on your own tasks, because a benchmark ranking from the vendor does not prove it fits your work.
The Verge · 28 Oct
Ars Technica: Gemini 4 Carbon is a small step up, and not yet usable (opens the source in a new tab)
Ars Technica calls Gemini 4 Carbon a slightly better version of Argon. Its headline says you can't use it yet.
Why it matters: Do not rebuild any workflow around Carbon until you can actually access it and check the price.
Ars Technica · 28 Oct
NVIDIA Nemotron 3.5 Transcribe and Granary v2 released (opens the source in a new tab)
NVIDIA released new multilingual transcription tools on Hugging Face. The title says you can prompt and adapt them.
Why it matters: If you transcribe calls or meetings in more than one language, this is worth a look as an alternative to paid services.
Hugging Face · 28 Oct
Tools and products
Microsoft introduces the Copilot harness for multi-model agents (opens the source in a new tab)
Microsoft introduced a Copilot harness that it says lets agentic work run across multiple AI models and surfaces. It says the hard parts are performance, quality and cost.
Why it matters: If your team uses Copilot, expect it to use more than one model behind the scenes, so watch for changes in quality and cost.
Microsoft · 28 Oct
HighLevel launches Client App Builder for branded mobile apps (opens the source in a new tab)
HighLevel says you can describe an app and AI will build a branded Android app. It connects to your existing Client Portal.
Why it matters: If you run an agency on HighLevel, you may be able to offer clients a mobile app without hiring a developer.
HighLevel · 28 Oct
HighLevel adds Vibe Coding to build client portals with AI (opens the source in a new tab)
HighLevel says you can build a custom Client Portal by describing what you want. It stays connected to your members, courses and communities.
Automation
HighLevel puts workflows into MCP for AI assistants (opens the source in a new tab)
HighLevel says Workflows in its LeadConnector MCP server, now in private beta, let Claude, ChatGPT, Cursor and other AI tools find, explain, build and publish workflows. They can also explain what failed and for whom.
Why it matters: If you run HighLevel, you may soon debug automations by asking an AI assistant, but it is a private beta so ask for access.
HighLevel · 28 Oct
HighLevel launches a Workflows API in private beta (opens the source in a new tab)
HighLevel says the new API can create, read, update, publish and delete workflows. It can also read execution logs and statistics.
Why it matters: If you have a developer, you could manage many client workflows from code instead of by hand.
HighLevel · 28 Oct
n8n explains how it moves its automation team beyond a service desk (opens the source in a new tab)
n8n's head of internal AI and automation shares four principles her team uses to help every team build and own its automations.
Enterprise
OpenAI and ServiceNow expand their partnership (opens the source in a new tab)
OpenAI says it will bring its models and agents into ServiceNow. ServiceNow workflows will also come into ChatGPT and Codex.
Why it matters: If you or your clients use ServiceNow, AI agents inside those workflows may become easier to set up.
OpenAI · 28 Oct
Anthropic launches tools to show what AI spending buys (opens the source in a new tab)
TechCrunch reports Anthropic launched tools that show companies how users spend money on its models. It comes ahead of a planned IPO.
Why it matters: If you pay for Claude across a team, you may be able to see which uses are worth the money.
TechCrunch · 28 Oct
Anthropic expands its GSA partnership to bring Claude to government (opens the source in a new tab)
Anthropic announced a wider partnership with the GSA to bring Claude to all levels of government.
Why it matters: If you sell to government or work with public bodies, expect Claude to show up more in their tools.
anthropic.com via Hacker News · 27 Oct
Research
Dayjob: Tax tests AI agents on 50 professional tax assignments (opens the source in a new tab)
Surge HQ published a test of AI agents on 50 professional tax assignments.
Why it matters: If you use AI for finance or tax tasks, results like these help you judge how far to trust it.
surgehq.ai via Hacker News · 28 Oct
Google Research proposes a Sleep framework for LLM learning (opens the source in a new tab)
Google Research describes a framework called Sleep that helps language models retain knowledge as they keep learning.
Why it matters: This is early research, but it points toward AI that learns from your business over time.
research.google via Hacker News · 28 Oct
MIT Technology Review on giving AI tools enterprise context (opens the source in a new tab)
MIT Technology Review says different AI tools in one business can produce wrong or confused results without shared context.
Why it matters: If you use several AI tools, make sure they work from the same business facts.
MIT Technology Review · 28 Oct
Also worth knowing
Report says Claude Code sends background agent transcripts to its model (opens the source in a new tab)
A post on GitHub Gists claims Claude Code regularly sends each background agent's transcript to its model. This comes from one writer, not from Anthropic.
Why it matters: If you use Claude Code with client or private data, check what gets sent before you rely on it.
gist.github.com via Hacker News · 29 Oct
Microsoft is ending Copilot support on smart TVs (opens the source in a new tab)
The Verge, citing Windows Central, reports Microsoft is ending Copilot on all smart TVs this year. It was not announced by LG or Samsung.
Why it matters: It shows AI features can be dropped, so avoid building a process on one that is not core to a product.
The Verge · 28 Oct
OpenAI essay looks at how AI might enter the org chart (opens the source in a new tab)
OpenAI published a piece on how AI might enter the org chart, beneath, beside and between people.
Why it matters: It is a useful prompt for thinking about which roles in your team AI could support.
OpenAI · 28 Oct