Skip to content

Do I Need a Developer to Make AI Work With My Business?

November 14, 2026

An owner trying to work out whether AI could help wrote this:

“Every time I try to understand it I feel like I am reading something meant for developers. I do not want to self host anything, mess with APIs, or build custom agents from scratch.”

We didn’t collect that thread ourselves. We read the quote in a round-up about AI tools and non-technical owners from May 2026, and that round-up, from a site that describes its articles as AI-assisted, doesn’t say where the original was posted. So take it as one owner’s words, reported second hand. The feeling in it is common enough that you probably recognize it.

The short answer: for almost everything that would save you time this month, you don’t need a developer and you don’t need an API key.

There is a point where that stops being true. It’s worth knowing exactly where it sits, because most of the jargon you’ve been reading lives on the far side of it and was never aimed at you.

What the owners getting value out of AI are actually using

Goldman Sachs surveys graduates of its small business program each year. In the August 2026 round, 2,473 of them answered. The report breaks down how the ones using AI work:

  • 86% say a standalone chat tool is a primary way they use it.
  • 54% name AI features built into software they already pay for.
  • 18% use agentic AI. 19% use coding tools. 12% use custom-built solutions.

One caveat before you read too much into that. These are graduates of a Goldman Sachs business education program, so they’re a selected and trained group. Their adoption runs ahead of small businesses generally.

The report’s own line about those last three figures: they “tend to require more technical investment and signal a deeper level of integration.”

So the answer is already sitting in somebody else’s survey table. Nearly all the value owners are collecting comes from a chat tool and from AI inside software they already had. Neither one gets built.

Cost, from the same survey: 73% of AI users spend under $100 a month on AI tools, and 16% use only free ones. The tools are cheap. Building is the expensive part.

The national picture is thinner, and it splits by headcount. The Census Bureau put AI use at 19.8% of businesses as of 3 May 2026. Fewer than 20% of firms with four or fewer employees reported using it. Between December 2025 and May 2026, use rose among firms with at least 20 employees and “didn’t change significantly” among firms with fewer than 20.

That survey doesn’t ask why, so nobody can tell you the cause from it. When owners are asked directly, they name the thing you’d expect. In an earlier Goldman Sachs survey of 1,256 owners in early 2026, 49% of AI users said a lack of technical expertise was a barrier, just behind data privacy and security at 50%.

Four rungs, and you should try them in order

Most of what gets called an AI project is one of four things. They are not equally hard, and almost nobody tells you to try the easy ones first.

Rung 1. The AI inside software you already pay for

This is where most real integration is happening, and the setup work is usually nothing.

Intuit announced a set of QuickBooks AI agents in July 2025. By Intuit’s description they handle bookkeeping and transaction categorization, draft personalized email replies to customers, and suggest and schedule meetings. Google added its AI to Workspace Business and Enterprise plans in January 2025 “without the need to purchase an add-on”, and said it was making pricing adjustments to those plans at the same time.

Read both as vendor descriptions of vendor products, because that’s what they are. The point holds anyway. If you pay for accounting software, email or a booking system, some of this is already switched on and already paid for. The work in front of you is learning it.

Rung 2. A chat tool, signed in to your own accounts

You can connect a chat tool to your own email, files or calendar without touching code. You sign in and click allow, the same way you’d connect any two apps you already use.

Anthropic’s own help pages draw the line neatly. Ready-made connectors get switched on from settings by signing in. For anything without one, “…developers can build and host remote MCP servers” (support.claude.com).

That sentence is the whole distinction. If a connector already exists for your system, connecting is a sign-in. If it doesn’t, somebody writes software.

Rung 3. A product built for the one job

A booking system that sends its own reminders. A phone answering service. A tool that asks your customers for reviews.

This is a subscription and a setup form. You configure it, and nobody builds anything. If one product does exactly the job you had in mind, buying it beats building, and it isn’t really an AI project at all.

Rung 4. Wiring two systems together yourself

This is the rung that bites, and it’s the one all the advice points you at.

The tools are genuinely sold as codeless. Zapier sells itself as a no-code automation platform that connects thousands of apps, with a free tier of 100 tasks a month and paid plans from $19.99 a month on annual billing for 750 (zapier.com/pricing).

The vendors are also clear that no-code isn’t the same as no-skill. n8n’s documentation points you to its cloud version if you lack technical expertise and don’t want to manage infrastructure, and says to self-host only if you have the infrastructure and technical resources for it (docs.n8n.io).

Which settles one worry straight away. The owner quoted at the top doesn’t want to self host anything. They shouldn’t. The vendor agrees with him.

So what is an API key, and when do I need one?

Plain English, once, so the next article you read makes sense.

API. The socket on the back of a piece of software that lets other software plug in. You never see it. The tool you’re using does.

Connector or integration. A plug that’s already been made for that socket. If one exists, connecting is signing in and clicking allow.

API key. A password that one piece of software uses to prove to another that it’s allowed in. Unlike your password, it’s a long string you paste into a settings box, and it usually has no second step protecting it. Google’s own documentation is blunt about what that means: “Unrestricted API keys are insecure.” (Google Cloud docs)

Webhook. An announcement one system shouts when something happens, like an invoice being paid, so another system can act on it.

Self-hosting. Running the software on a computer you’re responsible for. That includes the updates, the backups and the outage at 2am. Not a job for a business owner.

One rule worth holding on to. An API key is a credential, like a key to your premises. It belongs in an account with your name on it. If somebody sets one up for you, they should be creating it inside your account, not theirs.

Where a developer really is needed

Four cases, all documented by the vendors rather than by us.

  1. Self-hosting anything. n8n says so itself, in the guidance above.
  2. A connector for a system that doesn’t have one. Anthropic’s page says “…developers can build and host remote MCP servers.”
  3. Custom actions on a chat tool. OpenAI’s developer documentation describes the job as writing out the schema of an API call and setting up authentication so the tool can call the other app with your credentials. That’s programming, described as programming.
  4. Anything in that 12%. Custom-built work, in Goldman Sachs’s words, tends “to require more technical investment”.

Notice what isn’t on the list: a developer on your payroll. The US Bureau of Labor Statistics put median pay for software developers, quality assurance analysts and testers (one combined group) at $134,040 a year, or $64.44 an hour in its May 2025 wage data. For a two-van plumbing business, hiring is the wrong frame entirely.

What you need past rung 3 isn’t a programmer. It’s somebody who sets the thing up and owns it when it breaks.

Does it run while you’re watching?

That question splits this whole subject better than code versus no-code does.

Rungs 1 to 3 are attended. You’re in the loop. You see the output before anyone else does. The worst case is five wasted minutes.

Rung 4 and anything past it is unattended. It runs while you’re out on a job, it writes into your systems, and it can talk to your customers. That’s where API keys appear, and where the cost of a mistake stops being five minutes.

It’s also where failures go silent. A builder on the Make community forum published the failure list from a quote-chasing automation they put together in August 2026. The post lists seven things that cost them hours, all of which ran without a single error. They’re sharing a free template, so read it with that in mind. The failures are still worth your attention:

  • A mis-typed field mapping “silently produce[s] nothing”.
  • A missing space in an email search returns zero results, “indistinguishable from ‘no reply found’”.
  • A column mapping defaulted wrong, so values landed in the wrong places. In their case it overwrote a client’s name and email address.
  • A leftover copy of a spreadsheet ID meant the workflow “emails but never writes back, so it re-sends every day”.

The last two reach your customers. One had a client’s name and email address overwritten in the builder’s own records. The other emailed somebody every single day until a human noticed.

None of that throws an error, and it has very little to do with whether you can code. A non-developer on the n8n forum built a first workflow to send an email when a web page changed, then couldn’t tell whether it had worked: “I don’t get any test fails but I also don’t get an email when I test the send gmail node.” The tool didn’t refuse them. It did nothing, and they had no way to tell the difference.

Can’t I just get ChatGPT to build it?

Fair question, and it’s a real route. People are doing it.

In October 2026 an owner posted on the n8n forum about nearly three months building an automation platform with ChatGPT and Claude. Their words: “I have no formal coding background, no outside developer helping me, and no technical cofounder.” They were building something serious, pulling data out of PDFs that arrive by email and re-entering it into other systems. Where they had got to after three months: “I’m looking for help figuring out whether I’m on the right track, what needs to change, and whether paying someone is the best next step.”

That’s the honest version of this route. It works, it costs more time than anyone expects, and it can still land you back on the question you started with.

There’s a second cost you can’t see while you’re building. The security firm Veracode tested more than 100 language models across 80 coding tasks in 2025 and reported that “when given a choice between a secure and insecure method to write code, GenAI models chose the insecure option 45 percent of the time”. That’s a security vendor publishing research about a problem it sells tools for, and it measures professional code rather than an owner’s weekend automation. The 45% isn’t a number about your workflow. The direction is what matters, and the direction is that the code comes out looking fine either way.

If you pay somebody, ask for these six things

Another owner on the Make forum worked out the honest answer and went looking for help. They wanted two manual approval checkpoints “so nothing fires without me signing off first”, and documentation clean enough that they could run the thing themselves after handover. Then this line:

“I’m not technical but I know what I want it to do.”

That’s the position worth aiming for. You don’t have to learn the tool. You have to know the job and own the result.

So when you hire the build, ask for these.

  1. Every account and API key created in your name, in logins you control. Whoever builds it gets access, not ownership.
  2. A written description of what it does, in English, that you could hand to somebody else.
  3. A test you can run yourself that proves it’s still working.
  4. An alarm when it stops. The default failure mode is silence.
  5. The monthly running cost, itemized. Subscriptions, plus whatever the support arrangement costs.
  6. A named person and a response time for when it breaks on a Friday evening.

The test: can you do this task with yourself in the loop? Then no developer, and probably no new software. Does it have to run while nobody’s watching? Then somebody has to own it, and that somebody needs a name, a test and an alarm.

What we’d tell you to do first

Name the one task. Then name every system it touches. “Chase unpaid invoices” touches your accounting software and your email. That list of systems, not the AI, decides whether this is a sign-in job or a build job.

Then check what you already pay for. More than half the AI users in that Goldman Sachs survey get it from software they already had. Look there before you buy anything.

Then do it attended for a week. Open the chat tool, connect your own email or files by signing in, and run the task with you in the loop. If it doesn’t work with you watching, it won’t work unwatched. That week also tells you whether the task is worth automating before anybody spends money on it.

If one product does exactly that job, buy the product.

Only when it genuinely has to run unattended do you have a decision to make. Learn one no-code tool properly and accept that you now maintain software, or pay somebody to build it and keep it running. There is no third option where it runs itself.

There’s also help outside the vendor market, and it’s worth a call before a purchase. The SBA’s Small Business Development Centers provide “professional, high quality, individualized business advising and technical assistance” to existing businesses (sba.gov). A bill to make AI guidance an explicit part of that job, the AI for Main Street Act, passed the House and went to a Senate committee in January 2026. It isn’t law yet.

Where this leaves you

You were right that the articles are written for developers. You were also right that you don’t want to self host anything. Neither of those facts is standing in your way.

What stands in your way is one question about one task: does it have to run while you’re not looking? Answer that and you know whether you’re signing in or hiring.

Most of the time, you’re signing in.

Keep reading

Want to talk it through?

Tell me about the task that takes up your team's time.

Newsletter

Plain-language notes on software and AI for small business

One short email when there is something worth sharing. No spam. Unsubscribe any time.