AI Guides › Playbooks
By Nigel Guy · 6 min read
Headlines about Anthropic's valuation tend to produce one of two reactions: "I should buy into this" or "I should build the next big thing on it". Both skip the dull question of what you can do, for whom, that someone will pay for. A company being worth a lot says almost nothing about whether your small slice of its ecosystem will earn anything.
The rule: treat the growth numbers as evidence that demand exists, then pick one lane where you already have a skill, a customer and a way to prove the work. Nothing in this guide is investment advice or a promise of income.
Anthropic's own announcement of its Series H, dated 28 May 2026, states a $65 billion raise at a $965 billion post-money valuation, and says run-rate revenue had passed $47 billion earlier that month. Its February Series G was reported at $30 billion at $380 billion. Later reports put run-rate revenue above $65 billion by the end of July 2026; that figure comes from press and analyst sources, not from an Anthropic page I could check, so treat it as reported rather than confirmed.
Two cautions. First, "run-rate" means a recent period annualised, not revenue banked. Second, Sacra's company profile gives a revenue mix and labels its own figures as estimates from company statements and third-party sources, not audited accounts. It estimates roughly 70-75% of revenue comes from pay-per-token API use and 10-15% from subscriptions. If that is right, the money is mostly in businesses building Claude into their own work, which is where a freelancer or small agency can plausibly help.
I could not verify a Forbes article on this, so none of the figures here rely on one.
Draw four rows. For each, score yourself 0-2 on three columns: Skill (can you already do it), Buyer (can you name three real people or firms who would pay), Proof (can you show a working example within two weeks). A lane scoring under 4 out of 6 is a hobby, not a plan.
| Lane | What you sell | Why demand exists | Typical catch |
|---|---|---|---|
| 1. Implementation | Setting up Claude for a business: workflows, prompts, connectors, staff handover | Firms have licences and no one who has configured them | Needs a specific outcome, not "AI consulting" |
| 2. Building blocks | MCP servers, Skills, small tools that connect Claude to a system | MCP is an open standard for connecting AI apps to data and tools, so one build can work across clients | Free versions exist; you need a niche or hosting/support angle |
| 3. Enablement | Training, written guides, internal playbooks for a team | New features arrive faster than staff can learn them | Generic courses are everywhere; sell to one profession |
| 4. Vertical workflow | A repeatable process for one trade: quoting, case notes, compliance checks | Specialist context beats generic prompts | You need real domain knowledge and care over confidentiality |
Implementation. Anthropic launched the Claude Partner Network on 12 March 2026, committing $100 million for 2026. Its announcement says any organisation bringing Claude to market can join, that membership is free, and that partners get portal training, a directory listing and access to certifications. Its first certification, Claude Certified Architect, Foundations, is aimed at solution architects building production applications. Read the current terms before you pay for anything that claims to be "official". Large consultancies are anchor partners, so a solo operator should aim at small firms those giants will not serve.
Building blocks. The MCP site describes it as an open-source standard, with Claude and ChatGPT among the clients that support it. That helps you: one connector, many clients. It also means plenty of free ones already exist, so search before you build.
Enablement and vertical work. These need the least code and the most judgement. The cost of entry is a paid plan for your own testing. At time of writing Claude's pricing page lists Pro at $20 a month, or $17 a month billed annually (about £15 at time of writing), and Max from $100 a month (about £75 at time of writing); check the £ price at checkout, as tax and exchange rates change it.
Priya runs a three-person bookkeeping practice in Leeds. Her card: Lane 1 scores 3 (she has never configured a system), Lane 4 scores 5 (she knows year-end chasing, has twelve clients, and can show a before-and-after). She picks Lane 4 and writes: "I set up a client-chasing workflow for small bookkeepers so missing-records requests go out in minutes, not an afternoon." She tests it on two of her own clients, removes names, and uses the result as proof. Her kill date is 60 days. She doesn't build software or chase a partner badge. Whether it earns anything is unknown, and that is the point of the kill date.
You are a sober business adviser helping a UK freelancer choose where to focus when selling work that uses Claude.
My background: [YOUR_SKILLS_AND_EXPERIENCE]
People or firms I could approach: [YOUR_REAL_CONTACTS_OR_SECTOR]
Hours per week I can spare: [HOURS]
Budget for tools, in £: [BUDGET]
Task: score me 0-2 on Skill, Buyer and Proof for each of four lanes (implementation, building blocks, enablement, vertical workflow). Then:
1. Show a table with scores and a total out of 6.
2. Recommend at most one lane, or say none scores 4 or more.
3. Draft a one-line offer for it, with a measurable result.
4. Give a two-week plan for one proof piece.
5. Suggest a 60-day kill date and what evidence would justify continuing.
Rules: if any input is missing, ask me before scoring. Do not invent market sizes, prices, earnings or statistics. Mark every assumption as an assumption. Before answering, check that every score is backed by something I actually told you.
Fill in the four bracketed inputs; the more specific your contacts, the more useful the scoring.