Agent, artifact, app: what’s the difference and why it matters

An agent is the conversation. An artifact is the thing the conversation hands you. An app is a system that keeps working after you close the tab. Confusing one for another is the most common mistake teams make when they build a DIY tool themselves.

The usual version is a team that says “we built an app” when what they built was an artifact. It looked finished and it felt like software, but it had no way to keep working once the build was made. For the wider build-versus-buy question, see Can you build your own value selling tool with AI?

What is an AI agent, in plain terms?

An agent is the chat. You prompt it, it reasons through the exchange, and it answers. The intelligence lives inside the conversation, which is why the output can be genuinely good.

An agent’s memory is the thread you are in. Once the thread ends, you have a transcript: useful to reread, but not a running system. It can produce an excellent value case and then have no record of the deal the next time you open it.

What is an artifact, and what happens to it?

An artifact is the output the agent produces: the document, the calculator, the business case it wrote for you. It is a snapshot of one moment, fixed the instant it is made.

You can save it, send it, or drop it into a deck. What you cannot do is expect it to update itself when the deal changes. If the champion adjusts an assumption, the artifact does not know. If the customer moves the timeline, the artifact still shows the old numbers. It captured a point in time and stopped there, because keeping current was never part of what it does.

That is fine for a one-time asset. It becomes a problem when you needed a living value case and assumed you had one.

What makes something an actual app?

An app has a backend. It holds state, so today’s case reflects what changed yesterday, and every seller sees the same governed version. It connects to your CRM, keeps the math consistent as the underlying model shifts, and survives the person who built it moving on.

That backend is the difference, and it is invisible in a weekend build. When you watch someone generate a business case in Claude, you are watching the artifact get made. You are not seeing the infrastructure that would keep it accurate across a hundred deals, because there is none.

Why does confusing the three cost you?

Because a DIY tool can look like an app while it’s really just delivering an artifact, and the gap only becomes clear once the deal moves past the moment the agent captured.

A common sequence looks like this. A seller builds a strong case in a chat. Discovery continues over the following week. The champion reshapes two of the drivers, presales adds a proof point, and the AE learns the budget owner has changed. By the time all of that has happened, the original case is out of date and disconnected from every one of those changes, sitting in a closed tab while a new conversation starts over.

Across a team, the pattern compounds. Every deal lives in its own chat. Nothing rolls up, and nothing learned on the last deal reaches the next one. You end up with a collection of snapshots and no system.

Where does this leave a DIY tool?

As a strong one-time output that cannot hold a living plan across a sale.

That one-time output has real worth. A seller with a first version of the business case is in better shape than a seller with a hunch. The trouble starts only when you ask the artifact to do a platform’s job:

• Carry one value story from first call to renewal.
• Show a value leader that adoption is real.
• Feed the next deal with what this one taught you.

None of that is solved by a more capable chatbot. It is solved by a real backend, governed data, and a surface your buyer and your sellers can both work in. That is the difference between a value tool and a value platform, and we lay out the full ladder in What’s the difference between a value tool, a value platform, and a value function?

An agent is great for an outside-in, one-time, good-looking business case the buyer won’t believe and will ignore. If you want customer relationships built on the value you actually deliver, you need a platform.