- Cahn's Newsletter
- Posts
- "Shopping Wars Begin; Harvey Hits $8B" #Edition 32 (Nov 29- Dec 5)
"Shopping Wars Begin; Harvey Hits $8B" #Edition 32 (Nov 29- Dec 5)
This week, ChatGPT turns chat into checkout, Ironwood challenges Nvidia, DeepSeek V3.2 nears GPT-5 for less, Genesis targets a Manhattan Project for science.
Cahn Studios
December 06, 2025
The Big Conversation
Harvey's $8B Legal AI Valuation—The Enterprise Agent Economy is Here (Dec 4)
Legal AI startup Harvey confirmed a $160M Series F round led by Andreessen Horowitz, valuing the company at $8 billion—up from $5B in June and $3B in February. That's 2.67x growth in 10 months for a company founded in 2022.
Key metrics: 50+ of AmLaw 100 firms as customers, $100M+ ARR (as of August 2025), serving corporate legal teams at Fortune 500s. Harvey's platform automates document review, legal research, contract analysis, and workflow orchestration using Claude Opus 4.5 as its backbone. $160M Series F led by a16z; 50+ AmLaw 100 clients; $100M+ ARR. Automating legal research, document review, and contract analysis.
Why it matters:
The enterprise AI agent economy just crossed the inflection point. Harvey's $8B valuation signals that AI agents that replace human workflows (not just assist) are the next $100B+ market. Legal was first followed by healthcare, finance, and consulting.
The tension:
Shopping (ChatGPT) vs. Search (Google) vs. Social Commerce (Meta/TikTok) is now a 3-way AI battle. Whichever entity owns the "intent" layer wins the next decade of commerce. Read More
Elon Musk × Nikhil Kamath: a blunt, big-picture riff on money, markets, and machines—worth your whole commute.What's New this week
ChatGPT Shopping Research (Dec 1). Personalized buyer guides, product comparison, gift recommendations; Walmart + Target integration for instant checkout. Read More
DeepSeek V3.2 + Speciale (Dec 1). GPT-5-level reasoning at 1/10th API cost; Speciale variant matches Gemini 3 Pro on IMO/CMO/ICPC benchmarks. Open-weight release. Read More
Google Ironwood TPU (7th Gen) (Nov 29). 4x faster than Trillium; 192GB memory per chip; 9,216 TPU pod; Anthropic deploying 1M TPUs for Claude. Read More
Genesis Mission Executive Order (Nov 24). Department of Energy-led AI platform for scientific datasets; 60-day challenge identification; 270-day demo timeline. Read More
Canva Launches a "Creative Operating System":Canva rolled out a major platform refresh that includes Video 2.0 with prompt-based "Magic Video" and an AI "Design Model" that assists non-designers by understanding layout and visual balance.
AMD’s Infrastructure Validation:AMD's ecosystem partners Vultr and HPE announced major AI infrastructure moves, including a Vultr AI supercluster using 24,000 AMD Instinct MI355X GPUs and HPE's adoption of AMD's new "Helios" rack-scale AI platform, confirming growing real-world validation of AMD's AI accelerators.
Anthropic announced a partnership with Iceland's Ministry of Education and Children on the EdTech Innovation Hub, one of the world's first national-scale AI pilots for schools, giving teachers access to the Claude chatbot and related training resources.
Perplexity Patents Launched: Perplexity announced Perplexity Patents, an AI patent research agent that replaces traditional keyword-heavy patent databases with conversational search.
Fujitsu Develops Multi-Agent Collaboration: Fujitsu unveiled a new technology that allows agents to work together without sharing confidential data, primarily aimed at streamlining complex supply chain operations.
The Money Reality
December 2025 funding pace: $500M+ per day (extrapolated from first week)
Harvey's valuation trajectory: $3B (Feb) → $5B (June) → $8B (Dec) = 167% growth in 10 months
Legal AI market signal: Enterprise workflows that replace humans (not assist) command 3x+ multiples
Open-weight momentum: DeepSeek V3.2's GPT-5-level performance at open cost pressures closed-model pricing
Early December funding snapshot:
Category | Total Raised | Key Signal |
|---|---|---|
Enterprise AI Agents | $214M+ (Harvey, GetVocal, Flex) | Workflow automation = premium valuations |
AI Security | $28M | Deepfake defense is mission-critical |
AI Infrastructure | $2B+ (Google Ironwood deployment) | TPU vs GPU battle intensifies |
Biotech AI | $41.5M (AI Proteins) | Protein therapeutics = AI's next frontier |
Retail AI | N/A (ChatGPT Shopping) | E-commerce reimagined as conversational |
Founder tip: Add one line to your deck: Minutes saved, mistakes avoided, or money made per workflow. That’s your multiple.
Startup Radar — who to watch (and how to help)
Startup | One-liner | Stage | Why promising | What they need |
|---|---|---|---|---|
Model deployment from the cloud to the edge. | Series C | Superior Open-Weight Performance | Improved Multi-File & Edge Case Reliability | |
Real-time deepfake detection | Seed | Prevents AI impersonation at first contact | Enterprise adoption in finance/healthcare | |
Open-weight frontier models | Released | GPT-5 performance at 1/10th cost | Western enterprise partnerships | |
Accountable AI agents | Series A | Combines LLMs + deterministic logic | Customer service pilot deployments | |
AI-designed therapeutics | Series A | Protein design = post-LLM frontier | Clinical trial partnerships |
Cahn’s One Cent
Points to Ponder
Boards want ROI now: “AI for ROI” dashboards + agent-in-workflow beats lab demos.
Mix your stack: Open-weight where parity holds; closed for edge cases.
Price power shifts: Infra and “replace-not-assist” apps capture the fattest margins.
So what now?
Pick the needle: Choose one workflow that will move a core metric (a needle) fast—then build your first agent there, not everywhere. Try invoicing, lead-qual, onboarding, or support.
Wire it in (not next to it): intake → action → log → human check.
Measure with one line:
ROI = (minutes saved × runs/week × $/min) − infra cost.
Baseline 3 numbers → run the agent → re-measure.
Or revenue: ΔClosed-won or ΔCash-collected
If the graph doesn’t bend, it’s theater. If it does, you’ve got a business.
Bottom line: Agents that move money, minutes, or mistakes win. Pick one workflow, wire an agent into it (not next to it), and measure the delta.
Dev Corner: What to Build Next
Legal-adjacent agents: Harvey proved legal AI works. Build for: HR compliance, tax preparation, real estate contracts, insurance claims
Shopping orchestrators: Wire ChatGPT Shopping API into vertical SaaS (wellness product recommendations, spa retail upsells)
Deepfake defense: Use imper.ai's approach for customer verification layers (video consultations, telehealth, financial services)
Open-weight cost arbitrage: DeepSeek V3.2 at GPT-5 performance + 1/10th cost = rebuild your entire AI stack to cut OpEx 90%
Wellness-specific agents: Position as "Harvey for wellness" (booking, intake forms, post-visit follow-ups, product recommendations)
The Agentic Shift
Enterprise Agents Win:
Harvey's success proves that workflow replacement (not augmentation) is the $100B opportunity. Agents that autonomously handle 80%+ of a job category will dominate.
Shopping as Conversation:
ChatGPT Shopping kills comparison sites. "Find me a gift for my mom under $50" beats Googling + tab-switching 10x. Retailers integrating now own the next wave.
Open Models Closing the Gap:
DeepSeek V3.2 at GPT-5-level performance means closed-model pricing is under pressure. Expect OpenAI/Google to cut API costs 30-50% by Q1 2026.
Trust Layer Required:
imper.ai's $28M seed signals that identity verification is the missing infrastructure layer for AI agents. Deepfakes + impersonation attacks are the next cybersecurity crisis.
Government as AI Buyer:
Genesis Mission = DOE buying compute, datasets, and AI platforms. Startups that align with national priorities (energy, semiconductors, biotech) get government contracts.
Tools to Try
Tool | What it does | Best For |
|---|---|---|
AI for end-to-end on-brand campaigns | Founders, lean marketing teams, agencies | |
Slack-native AI for feature requests | Slack-first SaaS teams, PLG motion, high feedback volume, lightweight PM stack. | |
Open-weight GPT-5-level model | Cost-efficient AI workflows, reasoning tasks | |
Advanced AI image generation and editing model | Production-grade output of visuals by focusing on stability, consistency | |
Accountable AI customer service agents | Enterprise customer support automation |
Breakthrough Tools
1. Aha 2.0: An AI employee that automates the entire influencer marketing process—from expert matching and negotiation to contract management and performance tracking—allowing brands to simply review and approve campaigns. Try
2. ChatGPT Shopping Research: Personalized buyer guides + product comparison + instant checkout. The first AI-native e-commerce experience. Kills browser tabs.
3. LLM Council: A three-stage, multi-agent framework that forces different models (like GPT, Gemini, and Claude) to debate and peer-review each other's anonymized responses before a designated "Chairman" synthesizes the collective wisdom into one highly reliable final answer. Github link
4. Google Pomelli: AI marketing assistant that analyzes a business's website to automatically create a 'Business DNA' profile and generates instantly tailored, on-brand marketing campaigns, including copy and visuals, for small-to-medium-sized businesses (SMBs). Try
5. Imper.ai Platform: Real-time impersonation detection across Zoom/Teams/Slack. Analyzes device telemetry + network signals + behavioral patterns. The trust layer for AI agents. Check Out
AI x Creativity Exclusive
Storyboard Lock = Cleaner Cuts
Kling 2.x Start/End Frames let you lock opening/closing beats so shots stitch without morphs.
Why it matters: tighter pacing → higher retention.
Do this now (8 min): write a 3–5 shot list → set start/end frames per shot → render → assemble → watch the hook rate.
Thumbnail Factory v2 (Nano Banana)
Use a 6-image style board (logo, palette, product/faces) and batch 10–12 covers per video.
Why it matters: consistent look, faster A/Bs, fewer duds.
Do this now (10 min): export two top variants → A/B for 24h → keep the winner as your base style.
Layered Edits at Speed
Canva OS + Adobe Express AI Assistant = editable layers, quick swaps, on-brand templates.
Why it matters: one source → reels, carousels, shorts in minutes.
Do this now (9 min): import your brand kit → generate 3 post variations → schedule the best one.
Quick Plays
For Wellness Brands:
Use DeepSeek V3.2 to generate personalized wellness plans (nutrition, meditation, spa recommendations) at 1/10th the cost of GPT-5. Position as "AI wellness concierge."
For Local Spas:
Build a ChatGPT-style booking agent: "Book me a 90-minute hot stone massage with Sarah next Tuesday afternoon." Wire it to your scheduling system. Charge $99/month to competitors as SaaS.
For Social Media Managers:
Use ChatGPT Shopping API to create "product recommendation carousels" for Instagram. Auto-generate "gift guides" for wellness products. Monetize via affiliate links.
For Developers:
Rebuild your AI stack with DeepSeek V3.2. Cut API costs 90%. Use savings to build vertical agents (wellness, spa, fitness, nutrition).
Not to Miss Events
AWS re:Invent 2025 (Dec 1-5, Las Vegas) — Cloud + AI infrastructure, Ironwood TPU deep dives. Revisit
NeurIPS 2025 (Dec 9-15, Vancouver) — DeepSeek V3.2 technical deep dive, open-weight model trends. Register
The AI Summit NYC (Dec 2025) — Enterprise AI agents, vertical automation strategies. Save spot
Fireside Chat
If chat is the store and agents are the staff, what one workflow would you let an agent own end-to-end next week—and which metric (money, minutes, or mistakes) would prove it worked?
Would you trade 20% creative control for 5× output this quarter? Why/why not?
With open models near parity, which part of your stack goes open-first by January?
AI PUN
Modern Boss
Cahn's POV
Your unfair advantage is moving now. Build vertical agents (booking, customer service, product recommendations) using open models (DeepSeek) before your competitors do.
The stack is ready. The playbook (Harvey) is proven. The cost barrier (DeepSeek) is gone.
Start building. The enterprise agent economy waits for no one.
That's it for today!
Before you go we’d love to know what you thought of today's newsletter to help us improve Cahn’s AI Canvas experience for you.
⭐️⭐️⭐️⭐️⭐️ Nailed it
⭐️⭐️⭐️ Average
⭐️ Fail
See you Next Week,
Aditi and Swati — The humans behind Cahn’s AI Canvas.
📩This week felt like AI stopped asking for permission and started filing expense reports. Agents got gutsier, platforms picked sides, and chat stopped being “just chat.”
If you're a creator, dev, or artist trying to figure out where you fit—you're not alone. This isn't doom and gloom. It's about knowing what's real, what's hype, and where the actual opportunities are.
Want our setup checklist for local AI agents + our best prompt pack? Reply "Loop."
Stay Creative. Stay Updated. Get in Touch : [email protected]
Edition #32 covered Nov 29 - Dec 5, 2025. All news verified from mainstream sources with direct article links provided.
Disclaimer: The information presented in this newsletter is curated from public sources on the internet. All content is for informational purposes only.