{{first_name | Leader}}, welcome back. Today’s updates:
Early startups get one year of Framer’s Launch Plan free, limited spots available*
Anthropic launches Claude Sonnet 4.5
OpenAI’s agentic shopping automates personalized purchases
DeepSeek releases ‘sparse attention’ model
Tools, resources, and a prompt to analyse your competitor’s new launch ⬇️
Anthropic Launches Claude Sonnet 4.5
I’ve been tracking Anthropic closely, and their new Claude Sonnet 4.5 feels like a meaningful leap. It’s built for coding, workflow agents, and long autonomous runs, pushing the model closer to production-ready use in enterprise settings.
Key updates:
Extended autonomy: Runs for ~30 hours vs ~7 before, opening the door for longer coding and automation sessions.
Stronger coding reliability: Achieves top scores on benchmarks like SWE-bench Verified (per l), outperforming other leading coding models, including GPT-5 Codex and Gemini 2.5 Pro.
Better tooling: Ships with checkpoints, a VS Code extension, improved terminal interfaces, and context editing in the API.
Agent SDK: Let developers build on the same agent infrastructure Anthropic uses internally.
Enterprise reach: Available on AWS Bedrock, Google Vertex, Snowflake Cortex, Vercel, and in preview on GitHub Copilot.
Things to keep in mind
Rollout gaps: Some integrations (like GitHub Copilot) are still in preview.
Security validation: Prompt-injection defenses and alignment updates haven’t been independently tested at scale.
This update pushes Claude deeper into the enterprise developer stack. It reduces the fragility of long-running agents and gives teams better tools to integrate AI into workflows. But capacity alone won’t drive impact; how teams structure these agents will.
In other news
OpenAI’s new agentic shopping system automates and personalizes purchase journeys. It signals a shift in how digital commerce and procurement will be run, with direct implications for marketing, sales, and enterprise tech teams.
DeepSeek V3.2-Exp introduces sparse attention that cuts API costs for long-context tasks in half. This makes enterprise use of advanced language models for analytics, sales, and operations more affordable without sacrificing quality.
Framer is the design-first web builder that helps startups launch fast. Join hundreds of YC founders already building on Framer.
Who's Eligible?
Pre-seed and seed-stage startups.
Offer available to new Framer users only.
What You Get:
✅ 1 year of Framer's Launch Plan ($900 value)
✅ Fast, professional site — no dev team needed
✅ Scales with your startup — from MVP to full site
Spots are limited — apply today (last date) to claim your free year.
Productivity Tools
AlmReply – Drafts polished emails in seconds, adapting tone so you save hours on communication.
Simplescrapper – Pulls structured data from sites and feeds it into custom GPTs for analysis.
AIVA – Composes music in 250+ styles, opening possibilities for creative industries.
Lucidchart – Builds intelligent diagrams that help teams map workflows, align decisions, and plan effectively.
Get featured tomorrow: How do you use AI for business/personally? Interesting stories will be shared with 100K curious readers.
Useful Resources
A practical guide to protecting sensitive information across large organizations.
Insights into choosing and implementing software that scales with complex operations.
Details on the Chinese giant’s latest model, Alibaba’s Wan 2.5 AI, and where it competes globally.
A framework for assigning business value to predictive systems, often overlooked in deployments.
Benchmarks show Gemini 2.5 Flash Lite as the fastest proprietary model, with 50% fewer tokens per output.
💰 Funding
Datavault AI raises $150M to grow its data intelligence business.
Anything raises $11M at $100M valuation for AI app builder platform.
💼 Roles in AI
🐦 $3M/year: You + AI
How to make $3M/year solo with AI from anywhere
— #Aadit Sheth (#@aaditsh)
8:59 PM • Sep 29, 2025
AI ROI Forecaster
When to use this?
You need to see how an AI initiative will drive measurable business value before making an investment.
You are my Director of AI. I will describe an AI initiative below. Your job:
Identify the 3 most valuable business outcomes this initiative could deliver (e.g., cost savings, revenue growth, customer experience).
Estimate time-to-impact (quick wins vs. long-term gains).
Flag dependencies (data quality, infra, regulatory, change management).
Suggest one KPI to track per outcome.
Context: We are a [insert company details: industry, size, priorities]
Initiative: [paste your AI project idea or workflow here]
Output format:
| Outcome | Time-to-Impact | Dependency | KPI |P.S. Get more such prompts in the Prompting Playbook (free for you)
Q: Which AI model are you most curious to test/have been testing in 2025?
Stay curious, {{first_name | leaders}}
PS. If you missed yesterday’s issue, you can find it here.