{{first_name | Hey}}, welcome back.
Today’s issue is about Anthropic slashing the cost of small AI models, Microsoft CEO Satya Nadella warning companies to treat every AI model as compromised, and Google launching one agent to handle work across your apps.
These are today's updates.
Claude Haiku 5.5 cuts small-model costs by up to 90%
Nadella says to assume every AI model is compromised
Google launches one Gemini agent to handle work across your apps
Tools, jobs, resources, and last issue’s winning prompt ⬇️
Spotlight: ElevenLabs
ElevenLabs is heading into a busy stretch this month, with a mix of product education, enterprise events and startup activity on its calendar.
The company has been expanding rapidly in recent weeks. It launched in Singapore on October 8, following new launches in the Netherlands and Belgium, while its latest Eleven v4 and Eleven v4 Turbo models are now available across its platform.
Here are the main ElevenLabs events to watch over the next two weeks.
Oct. 20–21: Customer Impact Summit 2026, Cologne
Oct. 21: Startup Grant Demo Day
Neatprompts is where enterprise teams discover what's worth paying attention to in AI. Every week, 100,000+ readers rely on us for actionable insights and practical workflows.
If your product helps businesses adopt, deploy, or scale AI, we'd love to introduce it to our audience.
Top News
Anthropic cut the price of its small model with Claude Haiku 5.5. Anthropic says it costs about 75% less to run than Haiku 4.5 on average, 90% less for requests up to 100,000 tokens and 50% less above that.
Per-million-token prices for Claude Haiku 5.5, Haiku 4.5 and Sonnet 5.5. For prompts up to 100,000 tokens, Haiku 5.5 costs about 90% less than Haiku 4.5, Anthropic says. Table redrawn from Anthropic’s Claude Haiku 5.5 page.
Haiku 5.5 has an updated tokenizer, similar to Sonnet 5.5’s and Opus 5.5’s, so it uses slightly more tokens per task than Haiku 4.5. It is built for tasks such as classification, summarization and extraction, including live support, voice agents and in-app assistants. Anthropic says Sonnet 5.5 and Opus 5.5 are better suited to complex agentic coding.
It is available now on Amazon Web Services, Google Cloud and Microsoft Foundry. On the Claude Platform, developers can start with claude-haiku-5-5 model directly. Here’s Anthropic’s migration guide for more details.
Microsoft CEO Satya Nadella posted on X on Saturday that companies should “assume a model is compromised and contain it from the start,” calling it an “emergency brake.” He argued against treating AI as nested black boxes whose answers people simply accept or reject.
Nadella also proposed separating the model from the software that orchestrates it, with safeguards outside the model. Every significant action should be logged as tamper-proof, human-readable evidence, and an authorized person should be able to pause or shut a model down mid-task. Google’s Gemini agent, announced Thursday, lists the same pieces: a cryptographically attested identity per agent, audit trails attributed to the agent and an Agent Gateway firewall.
This comes on the heels of several recent cases of AI companies acknowledging they lost control of their models.
Google launched the Gemini agent on Thursday at Gemini at Work 2026, a single agent that answers questions, handles knowledge work, creates images and media, and writes and runs code, in the cloud even after you close your laptop.
It works inline in Workspace across Gmail, Drive, Docs, Slides, Sheets, Chat and Calendar with the same memory, skills and controls it has elsewhere. It also runs in Microsoft 365 and Slack. Agents get an identity, permission controls, a sandbox and a network gateway. Smart Routing matches each job to a model, currently Gemini or Anthropic’s Claude. Admins can set real-time spend caps that pause a project’s agent at the limit. Google lists customer results, including Ryanair at 35,000 employees and BNP Paribas at 65,000.
The agent is in private preview, with wider access coming to select Workspace Business and Enterprise plans, and Gemini Enterprise Plus has a free 30-day trial. Pricing is not announced yet.
Signals
Tools
Claude Managed Agents: A suite of APIs for building and deploying agents at scale, defining tools, personas, & capabilities
Microsoft-Decision-1: New model for fast decision-scoring, available in Microsoft Foundry through OpenRouter
Kōdlo: End-to-end sales workspace with unified context
Resources
Poll
Choose the prompt you want us to publish next
Market
Funding
Supabase raised $150M to build infrastructure for AI-generated databases
Oxide raised $445 million to expand manufacturing of its integrated, on-premises cloud systems
Roles in AI
Research Engineer, Cyber Gemini, DeepMind at DeepMind
Account Based Marketing Lead at Scale AI
Tutorial
Most people use Claude for answers and miss what it can do beyond chat. This video walks through 10 tips, from setting up memory and automating repetitive tasks to managing your inbox, organizing files, and building apps without coding. Watch to learn how to make Claude work more like a personal assistant and free up your time.
Prompt of the Day
AI Change Management Planner
When to use this?
Use this prompt when rolling out a new AI tool, redesigning workflows around AI agents, or helping teams move from AI experimentation to everyday use. It helps you turn a technology launch into a practical adoption plan, with clear owners, targeted training, employee communications, and measurable outcomes.
Act as an AI change management advisor who helps enterprise leaders introduce AI tools and workflows with minimal disruption and strong employee adoption.
Help me build a practical change management plan for an AI initiative, from preparing employees for the change to measuring adoption and business impact.
Start by asking me one question at a time about:
The AI initiative: What tool, system, or workflow is being introduced, and what problem should it solve?
The people affected: Which teams, roles, and how many employees will be impacted?
The current workflow: How do employees work today, and what will change?
The business goal: What outcomes are expected, such as time saved, better quality, faster decisions, or lower costs?
Employee readiness: How familiar are employees with AI, and what concerns or resistance might they have?
Leadership support: Who sponsors the initiative, and who will champion it within each team?
Implementation timeline: When will the initiative launch, and what milestones or deadlines matter?
Training and support: What training, documentation, and hands-on help are available?
Risks and constraints: Are there concerns about job security, data privacy, compliance, accuracy, or changes to responsibilities?
Success measures: How will we know employees are adopting the new workflow and getting value from it?
Once you have enough information, create a tailored... {Click on the button to view full prompt}Get more such prompts in the Prompting Playbook (free for you)
Stay curious, {{first_name | reader}}
PS. If you missed yesterday’s issue, you can find it here.
