The garage that passed the walkthrough
In a previous post, Buy vs. Build, Revisited, we argued that AI has changed the math. Building on Salesforce is now often the better call than buying a package. Buy the land, build the house.
That raises a fair follow-up question. If AI makes building this easy, why do you need a builder at all?
Every neighborhood has the answer parked at the end of a driveway. It's the DIY garage that looked fantastic in the photos. Then the first Wisconsin winter shows up. The door won't close, the roof sags under the snow, and when the house goes on the market, the inspector asks who pulled the permit. Nobody did.
Same nail gun, different builders
Hand a nail gun to a weekend DIYer and to a general contractor. Both will put up walls, and both will do it fast. The nail gun has no idea which of those walls holds up the roof.
AI is the nail gun. Anyone can now describe what they want in Salesforce and get back a flow, an Apex class, or a new set of fields in minutes. That's real, and it's good. We're not here to tell you to put the tool down.
But AI made the tools equal. It didn't make the builders equal. You'll get a house either way. The question is what kind.
One request, two prompts
Here's a request we hear all the time: when an Opportunity closes over $50K, email the account owner.
The DIY prompt:
Build a flow that emails the account owner when an opportunity is closed won over $50,000.
The contractor prompt:
Build a record-triggered, after-save flow on Opportunity. Fire only when a record is updated to meet the criteria, not every time it's edited. Store the threshold in custom metadata. Use an email template. Review the existing Opportunity automation first and fit into it. Make it bulk-safe, add a bypass for data loads, write a flow test, and deploy it through a sandbox.
Both prompts produce something that works in the demo. Both get a thumbs-up in the meeting. The difference shows up later.
| What happens when… | DIY build | Contractor build |
|---|---|---|
| You demo it | Works | Works |
| A rep fixes a typo on a closed deal | Sends the congrats email again | Stays quiet |
| Finance raises the threshold to $75K | Ticket to edit the flow | Admin updates one setting |
| Someone loads 10,000 records | Fires 10,000 times, or hits a limit | Bypassed, or handled in bulk |
| Another flow already runs on Opportunity | Two automations collide | Designed to fit alongside it |
| A new admin inherits it next year | Reverse-engineers it | Reads the test and the template |
Somewhere, a sales rep is on their fourteenth "Congrats on closing!" email for the same deal. They are starting to feel suspicious about the praise.
What the contractor knows that the nail gun doesn't
The contractor's prompt wasn't longer because they type more. It was longer because they've already seen what breaks.
Building codes. Governor limits, the security model, and sharing rules. These aren't suggestions, and AI won't apply them unless you ask.
Load-bearing walls. The data model. Put a field on the wrong object and every report, flow, and integration built on top of it inherits that mistake.
What's behind the drywall. The automation already running in the org. AI only sees what you show it. A contractor checks before they cut.
Inspections. Testing, including the ugly cases: bulk loads, bad data, the user with the weird profile.
Permits. Change management. A sandbox, source control, and a record of who changed what and why.
None of these show up in a demo, and all of them show up in year two.
This isn't an anti-AI post
We use AI every day, and it's a big part of why building now beats buying for so many of our clients. Our point is narrower than "be careful with AI."
The value has moved. It used to be in typing the solution. Now it's in knowing what to ask for, and in spotting what's wrong with what comes back. AI writes the flow in a minute. Experience tells you whether it should be a flow at all.
In our last post we argued that owning your solution makes it easier to maintain with AI. That's true, as long as the house was built to code. AI can maintain a well-built house. It will also happily patch a crooked one forever.
Before you move in
If someone on your team has used AI to build something in your org, good. Before you ship it, ask four questions:
What happens when this runs against 10,000 records instead of one?
What else already runs on this object, and how will they interact?
If the business rule changes next quarter, who changes it, and where?
Could someone who didn't build it understand it a year from now?
If you like the answers, great. If you don't, that's what a contractor is for.
Anyone can frame a house now. The question is whether you'd let your family sleep in it.
Salesforce is an ecosystem. Let's treat it as one, and build it to code.