Updated 29 September 2026 · by Amy Wilkinson
Tool guide
Claude for growth teams, beyond the chat box.
Most teams meet Claude as a chat window. The version that helps growth is the one that runs every morning, reads your CRM, drafts in your voice and waits for a person to say yes. Here is how the pieces fit.
The pieces
Five parts worth knowing.
Four current models
From Fable 5.1 for the hardest reasoning to Haiku 4.5 for fast, cheap steps. Most builds use two or three.
Cited research
Several searches that build on each other, across the web and connected sources such as Gmail and Google Docs, with citations you can check.
Repeatable work
Reads files, runs commands, and uses skills, hooks and scheduled tasks. The quickest way to turn a routine into something that runs.
Agents in your systems
The same agent loop as Claude Code, in Python or TypeScript, for agents that run inside your own infrastructure.
Reaching your tools
An open standard for connecting AI to tools and data. It is how Claude reads and writes to a CRM, Notion, Slack or Drive.
Models
Which model for which growth job.
Anthropic's own description of each model, and where we put it in growth work.
| Model | Anthropic says | Where it fits in growth work | API, per million tokens |
|---|---|---|---|
| Fable 5.1 | For demanding reasoning and long-horizon agentic work | Strategy passes, writing the ideal customer definition, reviewing a quarter of results | USD 10 in, 50 out |
| Opus 5.5 | For long-running agentic coding and knowledge work | Building and maintaining the system itself, account research, long drafts | USD 4 in, 20 out |
| Sonnet 5.5 | The best combination of speed and intelligence | Daily drafting, summaries, reply handling | USD 2 in, 10 out |
| Haiku 4.5 | The fastest model with near-frontier intelligence | Tagging, extraction and clean-up before a bigger model reads it | USD 1 in, 5 out |
From Anthropic's models overview, checked on 29 September 2026. Batch requests are half price. Fable 5.1, Opus 5.5 and Sonnet 5.5 take 1M tokens of context, Haiku 4.5 takes 200K. The third column is our own view.
A growth build on Claude
Six parts, whatever the system.
One place for the truth
Your offer, ideal customer, voice, rules and past decisions, in one place every step reads.
Connected, not copied
MCP connectors to the CRM, inbox, Notion, Drive and Slack. For content, the codebase, docs and call notes too.
On a schedule
Morning sourcing, a weekly content plan, a Friday report. The same job, run the same way, every time.
In your voice
Claude writes from the memory, with every claim tied to a source a person can check.
A person says yes
Drafts land where a named person approves them. The sending tool checks for that yes itself.
The next run knows
What was approved, rejected or replied to is written back, so tomorrow's run starts smarter.
The rules we build in
Guardrails live in the tools.
Do it yourself?
When Claude on its own is enough.
Plenty of founders get a long way with Claude, a few connectors and a skill or two. If your problem is one routine and you know the growth work well, build it yourself. Anthropic's docs are good, and a weekend goes further than you would think.
It gets harder with several routines sharing one memory, approval enforced in the tools, and keeping everything running when the model, your offer or your market changes. That is the part we build. See what we automate and what stays with a person, or the case study for what one system did in its first week.
Questions
Questions about Claude for growth.
Can Claude run marketing on its own?
What is MCP?
Do I need Claude Code to use Claude for growth?
Which Claude model should I use for marketing?
Do you build only on Claude?
Related
Start here
Let's build what runs without you.
Twenty minutes on how growth runs in your company today. If a build is not the answer, we will tell you what is.