You finished the course. You know what Cowork is, what it does, how skills work, what workflows it shines at. Now what?
The honest answer: not every AI problem in your business needs Cowork. Or any custom build. Sometimes the right move is to buy something off-the-shelf for $30 a month. Sometimes the right move is to do nothing at all. Most AI courses won't tell you that — they have one tool to sell.
Here's a framework that doesn't.
Four options for every AI decision
For any workflow in your business — customer follow-up, proposals, reports, lead qualification, anything — there are four possible answers. Most people only consider one.
Buy. An off-the-shelf solution already exists. You adopt it instead of building anything yourself. Sometimes it's a paid subscription (Klaviyo for email, NiceJob for reviews, QuickBooks for bookkeeping). Sometimes it comes free or bundled with something you already pay for — Anthropic's Claude for Small Business ships 15 pre-built workflows you install rather than purchase separately, included with paid Claude plans. The key is someone else built it, you adopt it. If the workflow is standard and an off-the-shelf option fits your stack, Buy is the right answer.
Bridge. Off-the-shelf software exists, but doesn't quite fit — or you need to connect two tools that don't talk to each other. A thin AI layer on top makes the existing software work for you. Cheaper than custom, more flexible than vanilla.
Build. The workflow is unique to how you run your business. No SaaS fits cleanly. The work touches files, schedules, multiple systems. This is where Cowork lives. This is where the rest of this course matters.
Skip. The workflow is low-volume, unstable, or just doesn't matter enough. Automating it costs more than just doing it. Leave it alone; revisit in 12 months.
The default in the wider AI conversation is "Build everything custom." That's wrong for most workflows. Most workflows are Buy or Skip. Knowing the difference is the entire skill.
This isn't new thinking. Joel Spolsky said it twenty years ago in "Strategy Letter V": "Smart companies try to commoditize their products' complements." In plain English: differentiate where you uniquely add value; buy off-the-shelf for everything else. Anthropic's own engineering team says the same thing about AI specifically — "finding the simplest solution possible, and only increasing complexity when needed. This might mean not building agentic systems at all." What we add here is explicit names for two underused options — Bridge and Skip — and a diagnostic an SMB owner can run in twenty minutes without an MBA.
Six questions to find the right answer
Before you decide to automate anything, run it through these six. Each one nudges you toward one of the four options.
- How often does this happen? More than 2x/week → consider automation. Less than 2x/month → likely Skip.
- Is it the same every time, or different every time? Same → a SaaS probably does it. Lean Buy. Different → no SaaS will fit. Lean Build.
- How many systems does it touch? One → Buy likely wins. Three or more → Bridge or Build.
- Would two different businesses in your industry do this the same way? Yes → commodity. Buy. No → unique to you. Build.
- If this was automated tomorrow and worked perfectly, what would change in your business? "I'd grow revenue / land more clients / save 5 strategic hours a week" → Build is justified. "I'd have more free time" → maybe Skip.
- What have you already tried? "Three SaaS, none fit" → the gap is real. Build. "I haven't tried any" → start with Buy. Build only after you know what doesn't work.
Two worked examples
A Calgary plumber wants to follow up with old customers
800 past customers. The plumber wants to send a "we're still here, $50 off your next service call" message.
Run the questions. Same every time? Yes. One system (the customer list)? Yes. Would two plumbers in town do this the same way? Yes. What would change if perfect? Some revenue from reactivated customers — useful, not strategic.
Answer: Buy. Mailchimp does this. Don't build it in Cowork. Subscribe, run a one-time campaign, move on. Save your AI energy for problems that actually need it.
The same plumber wants a Friday wrap-up of the week's jobs
Every job has notes in three places: paper invoice scans in /Receipts, customer notes in QuickBooks, follow-up reminders in Calendar. The plumber wants a one-page Friday summary: jobs done, revenue, follow-ups due, supplies running low.
Run the questions. Same every time? Yes. One system? No — three. SaaS that does this? No vendor stitches paper invoices + QuickBooks + Calendar into one weekly narrative. What would change if perfect? The plumber actually knows what's happening in their business each week, instead of finding out at year-end.
Answer: Build. This is exactly the kind of workflow Cowork was built for. Schedule it, point it at the three folders, get the Friday doc on the desktop by 5pm.
Same business. Two workflows. Two different answers. The framework discriminates.
Cowork's Skip signals
Even when the framework points to Build, Cowork specifically has three situations where it's the wrong tool. Spot them early or you'll waste time.
Your laptop closes at 5pm and goes home with you. Cowork's scheduled tasks only run while the desktop app is open and the computer is awake. If your work pattern doesn't support a computer left awake on a desk, scheduled tasks won't fire reliably. Change the setup (keep a dedicated computer on, or use a different kind of tool) or Skip.
You're in a regulated industry. Anthropic's own safety guidance is unambiguous: "Do not use Cowork for regulated workloads." Cowork activity is excluded from audit logs and compliance APIs, and as of May 2026, Cowork is also not available on HIPAA-ready Enterprise plans — only Claude chat is covered.
A reasonable question: Anthropic just shipped Claude for Small Business with workflows like Contract Reviewer and Tax-Season Organizer. Doesn't that mean Cowork is fine for legal and accounting practices? Not for client-protected work. Those workflows are useful for an SMB owner's own back-office — your own vendor contracts, your own books — but anything subject to HIPAA, attorney-client privilege, SEC/FINRA, or similar audit requirements still needs to live in Claude chat on Enterprise with a Business Associate Agreement, not in Cowork.
Until Anthropic closes the audit gap for Cowork, the BBBS verdict for regulated client work is: Skip Cowork for that work, Buy Claude chat on Enterprise with BAA for the parts that need compliance coverage.
You need memory that persists across standalone sessions. Cowork's memory works within Projects but not across standalone Cowork sessions. If your work can be scoped inside a Project, you're fine. If you need cross-session memory and don't want to think about Project structure, Skip until that changes.
These aren't reasons to dismiss Cowork. They're reasons to know when for you the answer is something else. That honesty is the framework's job.
What to do next
You now have:
- A framework for any AI decision in your business
- A diagnostic to apply it
- An honest read on where Cowork helps and where it doesn't
If you want to apply this to your actual business — surface the 5–10 workflows that matter most, route each through Buy / Bridge / Build / Skip, walk away with a roadmap — that's exactly what the Workflow Audit does. Two hours of your time, one page of recommendations. [Link to Workflow Audit booking.]
If you'd rather DIY: pick the workflow you do most often this week. Run the six questions. Pick the right option. Ship that one. Then the next.
One last note: BBBS is a decision discipline — it tells you what to automate and which path to choose. It doesn't tell you how to work with AI well once you've decided — how to write a prompt, judge what comes back, iterate. That's a different practice. Anthropic's own AI Fluency for Small Business curriculum teaches it as the 4D Framework — Delegation, Description, Discernment, Diligence. The two pair: BBBS for the decision; 4D for the interaction. If you want the next step on the interaction side, that's where to go.
The discipline isn't AI. It's deciding.