Apple · PlayStation · Charles Schwab
About Duncan Rogoff
Duncan Rogoff is the co-founder of Claude Code Club, the #1 community for learning Claude Code. He's an operator who built brands and products at Apple, PlayStation, and Charles Schwab before going all-in on AI.
In just 90 days he built a $100K+ AI agency, and he now runs The Build Room while sharing what actually works with an audience of more than 200,000 followers.
Inside Claude Code Club, Duncan leads the hands-on building - showing members how to ship real tools, agents, and automations with Claude Code, and get paid for them.
Areas of expertise
Articles
Claude Code can now draft a bug report about a session on its own and hand it to you for review, a tool called SendFeedback. Here's exactly what goes into that draft, why transcript sharing starts off by default, and the one check worth doing before you approve one on a session that touched client work.
Anthropic's temporary capacity boost on Claude Code ends September 14, replaced by a smaller permanent increase. Both the +25% and the -17% numbers you'll see floating around are true at once - here's the actual math and what I'm changing in every proposal this month.
Unlimited revisions sounds generous until it turns your calendar into a client's open queue. Here's the number I actually put in a proposal, the line that separates a revision from a new request, and how to say it without sounding like you're nickel-and-diming someone on day one.
Skip the deposit and you're financing a stranger's project with your own time. Here's the range I actually charge, the split I use once a project gets big enough to need one, and the one line that keeps a deposit from turning into an awkward conversation.
The upsell that actually lands isn't a new pitch, it's a specific next problem you already found while doing the first job. Here's the Second Ask, the one habit that turns a finished project into a second invoice without ever sounding like a salesperson.
Client work with Claude Code pays you now for a problem someone else already validated. A micro-SaaS pays you later, if at all, for a problem you have to validate yourself. The choice isn't about which one Claude Code can build faster, it can build either. It's about who has to go find the buyer.
A client asking for 'self-hosted Claude Code' isn't asking for a checkbox. It's asking you to stand up and maintain infrastructure their compliance team wants, and that's scope your normal build rate was never priced to cover. Here's how I separate the two conversations.
Most agency case studies are a description of work with a logo on top. A case study that closes deals carries three numbers, one client sentence, and enough detail that the reader recognises their own problem in the first paragraph.
Free scoping is the most expensive thing most builders do. A paid discovery phase gets you paid for the hardest thinking in the project, filters out the people who were never going to buy, and makes the build quote defensible.
The disclosure is almost never the problem. The timing is. Told in the proposal, it reads as your method. Discovered after delivery, it reads as something you hid, and the client reprices everything you already shipped.
AI made the building fast and left everything else exactly as slow as it was. That gap is why builders quote two weeks and deliver in six. Here is the estimating method I use, and the phases that actually consume the calendar.
The kickoff call is where a signed project either gets its rails or starts drifting. Most builders use it to be friendly and then discover in week three that nobody agreed what done means. Here is the Four Locks agenda I use to close a kickoff with the project genuinely de-risked.
Most AI agency websites are a logo, a list of tools, and a contact form. That page cannot close anyone because it never says what you sell or what it costs. Here are the six sections a buyer actually reads, in the order they read them.
Gemini CLI is free to start and lives in your terminal. Claude Code is a desktop app that treats your project as something to be maintained, not just edited. The choice comes down to how much of your work is one-off versus ongoing.
Lovable gets you a working app in an afternoon. Claude Code gets you a codebase you can still change in a year. The honest comparison is not about speed, it is about what you are left holding when the build is done.
There are two different Claude code review products and people keep confusing them. Here is what the managed pull request reviewer does, what the local /code-review command does, and which one you should be using.
A Claude Code plugin bundles slash commands, MCP servers, hooks, and agent instructions into one shareable package, so a whole team or a whole client roster starts every project with the same setup instead of everyone reinventing it. Here is how I install one, and how I package my own agency's workflow as one.
The build is only half the job. If the client cannot run, change, or understand what you shipped after you leave, you have delivered a black box - and black boxes generate the kind of support calls that quietly eat your margin. Here is the handoff checklist I use so a Claude Code project keeps working after I do.
A client who has never used an AI coding tool pictures their whole codebase getting pasted into a public chatbot. It doesn't work that way, and the honest answer is about what you control, not a policy you're asking them to trust blindly. Here is exactly what I say.
A statusline is the one line at the bottom of Claude Code that can tell you the model, the branch, the context left, and the cost - without asking. Here is how I set mine up so I stop losing track of long sessions.
Writing a project brief for Claude Code before you open the app cuts build time in half. Here is the exact brief format I use for every client project and internal build.
Building a mobile app used to mean learning Swift, Kotlin, or fighting a complex React Native setup. With Claude Code and Expo, you describe the app, Claude Code scaffolds the project and writes the components, and Expo handles the iOS and Android build. Here is the exact workflow I use to go from idea to a working app on a device.
Throwing an error at Claude and hoping it fixes it is not a debugging workflow. Here is the three-step method that actually works: describe what you expected, paste the full error, and tell Claude what you already tried. Thirty seconds of setup cuts debugging time in half.
Claude Code and GitHub Copilot solve different problems. Copilot autocompletes inside your editor; Claude Code thinks through entire builds end-to-end. Here is an honest look at what each one is actually for, where each wins, and which to use if you are a non-technical builder or agency owner.
I used to send AI project proposals that were five pages of detail and they closed slower than my one-pagers. Here's the exact one-page structure I use now - what goes on it, what stays off, and why a shorter proposal wins the deal more often.
Charging by the hour punished me for getting faster with Claude Code. The better I got, the less I earned for the same result. Here's why I switched to pricing the outcome, how I made the transition without scaring clients off, and the honest tradeoffs.
When a client asks you to guarantee results, saying yes is a trap and saying no loses the deal. The move is to guarantee the one thing you actually control - the work - and stop pretending you control the things you do not. Here is the Process Guarantee I use to answer with confidence and still be honest.
When a client asks you to drop the price, they are not telling you it costs too much - they are telling you they cannot yet see what they are getting. Discounting on the spot trains them to push again. Here is the Value-Anchor response I use to hold my price without losing the deal.
Most builders quote too early, on too little information, and then spend the project fixing a scope they got wrong in a message thread. A proper discovery call for AI projects takes thirty minutes and tells you the real outcome, the real budget, and whether this client is actually worth taking on.
The safe way to use Claude Code with Git is to commit before you let it touch anything, work on a branch, and review the diff before you accept. Here is the checkpoint habit I use so an AI editing my files can never cost me work I cannot get back.
Everyone selling AI services is shouting into the same LinkedIn feed. Here is the exact playbook I use to get AI clients on LinkedIn - a positioned profile, proof-led posts, and warm DMs that start a conversation instead of pitching - without cold-spamming a single person.
Everyone says niche down, nobody tells you how to choose. Here is the three-filter test I use to pick a niche for an AI agency - one that has real pain, real budget, and people you can actually reach - so you stop guessing and start booking calls.
You don't need a sales call to close an AI build. Most of my client work is closed over email - a clear scope, a single price, one deadline to decide, and a link to pay. Here's the exact email sequence I use to take a warm lead from interested to paid without ever getting on the phone.
Most AI cold emails get ignored because they brag about tools instead of naming a problem. Here is the exact four-line structure I use to turn a cold contact into a 15-minute call - and how to personalize it without spending an hour per email.
Non-technical clients don't buy AI - they buy outcomes. They don't care that you used Claude Code, an MCP server, or a subagent. They care that their invoices go out on time and their leads get answered in five minutes. Here's how to translate the tech into their language, demo the result, and price the outcome so the sale closes itself.
"Let me think about it" is where most builder deals quietly die. It is rarely about thinking. Here is how I find the real reason without pushing, and turn a soft maybe into a yes or a clean no.
Every prospect eventually says it: why pay you when they can just use ChatGPT? Here is the exact way I answer it in the room, why arguing about features loses, and the reframe that turns the objection into the close.
A CLAUDE.md file is the context Claude Code reads before it touches your project, and a good one is the difference between generic output and code that fits. Here is what belongs in a CLAUDE.md file, plus ready templates for web apps, client builds, automation, and content sites.
Cursor and Claude Code both put AI at the center of building, but they make opposite bets on where you work. An honest comparison of the AI editor and the terminal agent, and how to tell which one fits the way you build.
Learning to read client red flags before you sign is the cheapest skill in your business. Use the Green Light Test to spot the warning signs early and only take on clients worth having.
AI on client work is normal now - but there is a right way and a fast way to lose a client. A clear, save-worthy list of the do's and don'ts that protect your quality, your reputation, and your relationship with the people paying you.
A practical playbook for landing your first paying clients: pick a niche you can speak to, build one undeniable proof asset, and run outreach that leads with a result instead of a pitch - the exact path that turns a new AI agency into a booked one.
Repeat clients cost nothing to acquire, come with built-in trust, and refer their networks. Here is the delivery system and follow-up habit stack that turns a one-off project into a long-term client relationship.
Most freelance deals die not because of a 'no' but because of silence. This guide covers the exact cadence, messages, and mindset you need to follow up with a client without feeling pushy - and without losing the deal to someone who stayed in touch.
Knowing how to turn a one-off client into a monthly retainer is the difference between chasing new work every month and building a stable, compounding business. This post walks through the right moment to pitch, what to put in the offer, and how to have the conversation.
A Claude Code skill is a folder with one instruction file that teaches Claude a repeatable job it can reach for on its own. Here is how to build your first one from scratch, test it, and know when a skill is the right tool.
A client who won't pay is a process problem, not a personal one. Use the Calm Collection method to get paid without burning the relationship or your own peace of mind.
You can raise your rates without losing clients if you earn the raise first, give notice early, and anchor the new price to outcomes. This is the Value-First Raise I use.
A clear AI project contract protects both sides and prevents most client disputes before they start. These are the six clauses I never ship a project without.
Claude Fable 5 ran a two-month codebase migration in a single day in testing. That does not mean you point it at your legacy system and hope. Here is the four-phase method for migrating real code with Fable, keeping behavior intact and your client's trust with it.
Claude Fable 5 is built for long-horizon agents, the kind that plan, act, and keep going across millions of tokens. But the model is the easy part. Here is the contract every Fable 5 agent needs to actually finish the job instead of looping forever or quietly making things up.
A hands-on Claude Fable 5 review after putting the first public Mythos-class model through real work. What genuinely impressed me, where it frustrated me, and a straight answer on whether it is worth switching to right now, or whether you should wait.
Anthropic's new Claude Fable 5 ran a two-month codebase migration in a single day. For an AI agency, that is not a tooling update, it is a pricing question. Here is what changed, what to sell, and the client-work caveats nobody is talking about.
Knowing how to get client testimonials turns finished projects into your best sales asset. This is the simple ask I use to collect proof that actually converts.
Juggling several clients in Claude Code does not have to mean a dozen scattered windows. Here is the client cockpit method: one repo and one CLAUDE.md per client, one view to see every job, and the guardrails that keep client work from bleeding together.
Knowing how to handle scope creep is what separates a profitable agency from a busy one. This is the simple line I hold on every project so the extras get paid for.
SpaceX is buying Cursor's parent Anysphere for $60 billion in stock, days after its own IPO. Here is what the Cursor acquisition actually signals, and why the durable moat for an agency is the system you own, not the editor you open.
The right client discovery questions turn a guessing game into a confident quote. These are the ten I ask on every discovery call before a number exists.
To productize a service means turning custom, hourly work into a fixed, repeatable offer with a clear scope and price. Here is why it scales an agency, the CCC productization loop, and 3 productized offer templates you can launch this week.
Learning how to scope a Claude Code project is the difference between a profitable build and a quiet loss. This is the method I use to size work before a number ever leaves my mouth.
A client proposal template is the difference between winning a deal in an hour and losing a week. Here is the full reusable proposal structure we use, section by section, with a copy-paste Claude Code prompt to draft each part from your discovery-call notes.
Your agency AI tech stack is not 20 tools - it is a few layers that compound. Here is the CCC agency stack: build, automation, knowledge, delivery, and ops, with the trap to avoid at each layer.
AI automations for agencies are how lean teams grow margin without growing headcount. Here are 10 concrete automations you can build with Claude Code - what each one replaces and the rough trigger that fires it - plus the CCC automation ladder for deciding what to automate first.
Client onboarding is where most agencies quietly leak time and trust. This Claude Code client onboarding pack gives you 7 copy-paste prompts that turn discovery notes into a welcome doc, a scope summary, a kickoff questionnaire, an access checklist, and a first-week plan in minutes.
The best Claude Code plugins bundle commands, subagents, MCP servers, and hooks into one install so your agent shows up ready for a specific job. Here are the plugins worth installing, what each unlocks, and how to add one without bloating your setup.
The best Claude Code MCP servers are the ones that connect the agent to the tools and data you already use. Here are the ones worth installing, what each unlocks, and how to add them without bloating your setup.
Claude Code Artifacts now ships live, shared dashboards and interactive workspaces. Here is how an agency turns a one-off build into a deliverable the client can actually log into - and bill on a retainer.
A practical starter set of skills that earn their place on real work. Each one is described by what it does and when it pays off, so you install the ones that match your actual workload, not a pile you never use.
The June 15 Claude billing change turns programmatic Claude usage into a real per-seat cost. Here is how to reprice client work, claim your credit, and bill agents as infrastructure instead of eating the API bill.
A practical guide to scheduling recurring Claude Code agents - a morning report, a nightly cleanup, a weekly research pull - and the one rule that keeps unattended runs from costing you a client.
Four primitives, one decision. This is the plain-English guide to Claude Code skills, subagents, MCP, and hooks, with a table that tells you which to reach for and the mistakes that quietly slow your work down.
Claude Code can now write its own harness for a task, fan the work across subagents, and run it in the background while your session stays free. Here is how to turn dynamic workflows on and run your first one today.
Every business running Claude Code is bleeding context. They don't know it's a problem yet. The shops that show up with a productized memory-engineering deliverable are about to spend twelve months printing $5K retainers. Here's how I'm pricing it.
The client that pays the most isn't always the client that earns the most. I cut my biggest invoice in May, and the agency's margin doubled by the end of June. Here's the math that made the call obvious - and the playbook for spotting your version of the same client.
The fastest way to cap your agency's earnings is to charge by the hour for something Claude Code builds in a weekend. Here's the pricing system top AI shops actually use - and why the unit you sell matters more than the rate.
The real numbers behind my first freelance Claude Code invoice - what I charged, what I should have charged, and the pricing system I run my agency on now.
MCP servers are how Claude Code reaches out of the chat window and into your real tools - Gmail, GitHub, your database, the web. Here's what they actually are and why they change what you can build.
Learning Claude Code is fun, but turning it into income is the goal. Here are five concrete, repeatable workflows our members use to land clients and ship paid work.
Most builders learn the tools and then freeze at the part where someone actually pays them. Here's the 30-day plan I'd hand my younger self for going from "learning Claude Code" to "first paid client."
Built by
Claude Code Club is built and taught by David Iya and Duncan Rogoff - operators with two decades of shipping work across multi-billion-dollar brands.
Duncan RogoffApple · PlayStation · Charles Schwab
David IyaForbes 30 Under 30 · Y CombinatorOur team has worked with and been recognized by
Join Claude Code Club, the #1 community for learning claude code, for $9/month.
We use cookies to keep the site running, remember your preferences, and understand how people find us. See our Cookie Policy for details.
