Howdy y’all.
I spent a week in Austin, at sellers’ QBRs, and eating barbecue. I was enabling the field on Canva’s MCP and some of the workflows we can now position to our customers.
I’m always inspired by the engagement from our team. It truly energizes me, while... still being a ton of work.
Last week’s newsletter was very Clay-focused. This one is very much focused on Anthropic: the quarter, and this week’s long-form Business Breakdown of their new agentic buying experience. Then we flow into what we maybe missed in the world of AI this week.
On the menu this week:
🖥️ Anthropic’s reported $16.8B quarter, and a sales team that only takes complex deals.
🛠️ The play: where Claude Haiku 5.5 fits, and one task to test it on.
🔍 Growth teardown: Anthropic’s buying agent, and whether it’s the end of AEs.
📡 On the radar: how ChatGPT ads target buyers, a GPT-4o shutdown date and Google’s Gemini agent.
01 / THE BIG MOVE
🖥️ Anthropic’s reported $16.8B quarter
Anthropic made $16.8 billion in revenue last quarter, per Reuters, citing two people at its October 14 investor day.
The number that matters: about 45%. That’s the growth from the $11.6 billion Reuters reported for the second quarter. These came from attendees. There’s no filing.
What we know:
Profit: Reuters reports Anthropic was profitable on an adjusted basis in the quarter.
Where it’s heading: analysts who met executives that week estimated its annual run rate could reach $110 billion by year end, per Reuters.
The IPO: Reuters has reported the listing likely comes after the November midterms, at a target valuation above $2 trillion.
How it sells: Anthropic says it gets tens of thousands of inbound requests a month. Its contact sales page says the sales team “can only support complex deals, like 500+ seat deployments and BAAs.” Everyone else buys on their own or through a buying agent.
Why it matters for GTM: their buyers would still take a rep. Anthropic has so much inbound it has to be particular about what gets to one.
I ran six test chats with that buying agent this week for the long-form Business Breakdown. The short version is in 03.
02 / PUT IT TO WORK
🛠️ The play: move one step to a smaller model
In my adoption curve post, I said our harness, the software around the AI model, should spend smarter: a small model for the lookup, a stronger one for strategy. Claude Haiku 5.5 makes that split cheap.
Anthropic priced it to match OpenAI’s GPT-6 Luna to the cent: a tenth of Haiku 4.5, and about 5% of the mid tier, where Sonnet 5.5 and GPT-6.1 Sol both sit at $2 and $10.
List price per 1M tokens, Oct 16 | Input | Output |
|---|---|---|
Claude Haiku 5.5 (prompts up to 100k) | $0.10 | $0.50 |
OpenAI GPT-6 Luna | $0.10 | $0.50 |
Google Gemini 3.5 Flash-Lite | $0.30 | $2.50 |
Immediately, I think Haiku fits four jobs for us: process tasks in our CRM, fuzzy matching on accounts, simple categorization, and the simple data processing that enrichment at scale needs. We’ll likely move that work to this type of model.
Three personal workflows anyone can move to it:
Email replies: the routine ones. Keep the delicate ones on a bigger model.
Your assistant agents: the morning brief, the inbox sweep. First to move if you need to save credits.
Meeting notes: transcript in, summary and owners out.
TRY THIS ON ONE TASK
Move one scheduled agent to Haiku 5.5 for a week.
Pick the one that runs for you daily. Switch its model and read the output each day. No miss by Friday, leave it there and move the next one.
A daily brief that reads 50,000 tokens and writes 1,000 costs under a cent on Haiku 5.5 and about 11 cents on Sonnet 5.5. My arithmetic. Max and Team plans now include a monthly API credit, per Anthropic.
03 / GROWTH TEARDOWN
🔍 Anthropic launched a buying agent. Is this the end of AEs?
Anthropic now sells through a buying agent. Open its pricing page, click “Chat with buying specialist,” and a Claude agent runs discovery, quotes a plan and sends you to checkout. Below 500 seats, you never have to talk to a rep.
The agent works because checkout came first. 54% of Anthropic’s new enterprise logos in 2026 came through self-serve, Head of Industries Eleanor Dorfman told SaaStr in May. The agent can only close what self-serve can.
THE SHORT VERSION
Will this fully replace AEs? The answer: not really. The agent takes the buyers reps couldn’t reach. Reps keep the committees and the C-suite.
Where it hits first: low-ACV, high-velocity funnels. A low five-figure deal can close in a chat. A six-figure one still needs a person.
What I’d test: self-serve invoicing, so a deal that only needs an invoice never needs a rep.
04 / PIPELINE SIGNALS
📡 On the radar
404 Media ran its own ChatGPT ads to show how targeting works. Advertisers can add topic hints and upload customer lists of emails or phone numbers. OpenAI’s help center says targeting a list needs at least 25,000 matched users, and smaller lists can only exclude. So what: demand gen teams testing ChatGPT ads can keep current customers out today, and target a list once it matches 25,000 people.
OpenAI told developers on October 14 that GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini leave its API on January 14, 2027, fine-tuned versions included. It names gpt-6-luna as the replacement for all four. So what: if a workflow still calls one of these, it has to move anyway. Run the play above while you’re in there.
Google announced the Gemini agent on October 8. You give it objectives, and it works inside Gmail, Docs, Sheets and Calendar and connects to Salesforce, Slack and any MCP server. It picks between Gemini and Anthropic’s Claude models for each job, with spend caps per project. It’s in private preview for enterprise customers, per The Verge, and Google hasn’t said when it reaches the rest of Workspace or what it costs. So what: it makes the model split from the play above automatic. Workspace teams should ask their Google rep when it reaches their plan.
OVER TO YOU
💬 If reps were your bottleneck, who would get one?
For me, reps go where they drive incremental conversations, and the busy work goes to the agent. The hard part is defining what counts as busy work, and making sure the agent can actually complete it.
The other hard part is measurement. Running true growth experiments with humans in the loop is a challenge in itself, so the incremental impact of a rep is never a clean number. Who would get one on your team?
How was this week’s issue?
Robert
P.S. Know someone trying to put AI to work in go-to-market? Forward this to them. They can subscribe at bizgrowth.ai.