Buy the brain once. Use it in 20 different ways.
The build-or-buy conversation for AI tools is happening all around small businesses. Build means hiring a team and developing custom AI infrastructure. Buy means signing an enterprise software contract and waiting for implementation. Neither of those is actually on the table for many small businesses — and often that’s not a problem, because there’s a third option that might better fit how you actually operate.
You can configure. Take a capable general-purpose AI tool, shape it specifically around your business, and put it to work — without an engineering team, without a six-figure contract, and without a six-month procurement cycle.
Here’s the thing most AI coverage gets backwards: small businesses aren’t disadvantaged here. They’re structurally better positioned than enterprises to do this well. Let’s be specific about why.
Why small businesses have the edge
Pilot and production are often the same thing. At a large company, moving from “testing a tool” to “using it for real work” means navigating IT approval, security review, change management, and staged rollouts. At a small business, you decide on Tuesday and you’re using it on Wednesday. If it works, it stays. If it doesn’t, you change it on Thursday. Most enterprise AI projects spend months in “pilot” before anyone does real work with the tool. That feedback loop problem simply doesn’t exist for you.
You don’t have to navigate a web of approvals. That doesn’t mean you can go rogue — sensible data hygiene and a clear sense of what the tool has access to still matter. But the structural layers that slow large organizations down — IT reviews, compliance committees, governance deliberations — aren’t part of your process. You can move when something is ready, not when the calendar clears.
You know your own business. At a large company, the person who understands the workflow and the person who approves the tool are almost never the same person. At a small business they usually are. That matters enormously for AI, because the quality of what an AI produces is largely determined by the quality of context it receives. You can describe your customers, your tone, your constraints, and your edge cases — accurately, in detail — in a way no vendor can pre-configure for you. That proximity is what turns a generic tool into something that actually fits.
What configure looks like in practice
Configuration isn’t a one-time setup. It’s a gradual process — and you can start without any technical background.
1. Manage it well first
The first step is learning how to work with a general-purpose tool effectively. Think of it less like software configuration and more like onboarding a capable new hire who doesn’t know your business yet: be specific about what good looks like, what the output needs to accomplish, and what the tool shouldn’t do on its own.
Clear task decomposition — breaking down what you want into specific, well-scoped instructions — is the foundation. A tool like Claude Cowork is a good place to start. Learning what it can do at this level is itself the first configuration step.
You don’t buy a new app for every task involving tabular analysis. You buy a spreadsheet and learn how to use it. AI shouldn’t be treated any differently.
2. Systematize your business rules with skills
Once you know what a tool can do, the next step is making it remember. Most general-purpose agents — including Cowork — support skills: saved sets of instructions that can be automatically applied to recurring tasks.
If you review client deliverables the same way every time, that’s a skill. If every customer-facing email follows a particular structure and tone, that’s a skill. If you regularly produce a certain type of report, that’s a skill. You’re not writing code — you’re writing down the business logic you already carry in your head and making it available to the tool consistently.
This is where configuration starts to compound. Each skill narrows the gap between a generic AI response and one that actually fits how your business works.
3. Add connectors
The third step is linking the tool to the systems it needs to work with. MCP connectors (Model Context Protocol) are becoming standard across agent platforms and many business tools— they let your AI reach into a CRM, a calendar, a project management system, or a database and pull in live context without you manually copying things across.
Many connectors already exist and require nothing beyond authentication. If your CRM has an MCP integration, your agent can reference customer history directly. If your calendar is connected, scheduling tasks stop requiring you to be the relay. The ecosystem is growing fast — before building anything from scratch, it’s worth checking what connectors already exist for the tools you use.
If you want to build
Many small businesses may not need to build anything - at least not when they are getting started with AI. But if your use case is genuinely specific enough that configuration won’t get you there, the tools available are getting more accessible.
Managed agent frameworks let you define agent behavior, tool access, and decision logic without building raw infrastructure. Deep Agents, from the LangChain team, is one of the more capable options for teams that want to create multi-step agents with structured control flow — it’s a serious framework, but comes with documentation and is actively developed. Anthropic’s managed agent offering is in beta, but allows agent builders to build on a core agent brain (Claude), harness, and hosted environment that is configurable without bespoke engineering (you’ll still need to bring the control plane).
OpenClaw is also worth knowing about — it’s a free, open-source autonomous agent that runs locally on your machine and has built a large following fast. The ambition is real, but one of its own maintainers has noted that “if you can’t understand how to run a command line, this is far too dangerous of a project for you to use safely” — and independent security testing has flagged risks around third-party skills. For most small businesses, it sits firmly in build territory, not configure.
The honest caveat: even with a coding agent helping you build, you’re safest going down this path if you or someone on your team has a working understanding of AI product architecture, deployment, and security hardening — enough to review what’s being generated and recognize where it can go wrong. A coding agent accelerates the work meaningfully. It doesn’t replace the judgment.
Where to actually spend money buying
Not everything is just a question of configuration. There are cases where a purpose-built tool earns its price:
- The workflow is heavily regulated — financial, legal, healthcare — and the vendor’s compliance infrastructure is the actual value, not the AI layer on top.
- The product contains proprietary data or network effects you can’t replicate — a vertical SaaS built around an industry dataset or a marketplace that only works because everyone’s already in it.
- The integration complexity is high enough that a purpose-built tool has already solved the problem you’d spend months on.
The test isn’t “does it have an AI feature?” Most SaaS tools do now. The test is whether the core value of the product is something you couldn’t build with a good general tool and the right context.
We can help
If you’ve read this and want help getting started, we run Configuration Sprints that help you manage, systematize, and connect — your first skills, your first integrations, and the playbook to keep going. Book a 20-minute intro →
Josh is co-founder of Kynetyk, where he writes about AI, builds products at the intersection of AI and human experience, and helps companies design AI strategies that actually scale. Reach out at josh@kynetyk.ai.