Demo automation with AI agents

What if custom demo builds took 20 minutes instead of 2 hours? How would that change who you create custom environments for? Your sales velocity? What would you do with the extra time?

Proof of concept (POC) / trial

The biggest opportunity is to accelerate — and offload the manual work of — the discovery → demo turnaround.

Traditionally, you ask the prospect for their own data and workflows, upload and build that into your solution, and get the prospect into their own working account. It's high reward becuase prospects can see themselves in the system and you validate the functionality, but it's also high risk. It takes time and you could get their business model / data wrong and the "art of the possible" backfires when they object "that's not how we actually do it"

With agentic tools like Claude Code, you can significantly streamline the information gathering and demo environment build process, keeping the high reward intact while reducing the "takes time" downside. Here's an example workflow that I've seen work well:

1. Take the call transcript from your call recording software (e.g., Gong, Zoom). Best if you layer on real discovery notes and sales account research to ground the context and introduce higher-quality insights.

2. Drop the discovery text into your agentic coding tool (e.g., Claude Code, Codex). Have it extract the key pieces of information (like industry, company size, pain points) and summarize their key workflows and what they call things. This is essential for making things feel real by using their language and their terminology.

Example prompt

You're a solutions engineer for [X Company], who is building a custom demo environment for a prospect. I'll provide the call transcript and sales notes and you'll synthesize them into this format.
Industry and vertical
Pain points.
Top 3 reasons for engaging with us. What value are they hoping to get out of our solution.
Key stakeholders.
Who will be using the solution?
Key pieces of data that'll need to be in the solution. How will they move it over or integrate?
Key 3 workflows that they'd like to validate? These will be the demo flows that we'd like to show. Write these out in detail, using their language.

3. Generate realistic sample data (objects, records, users, scenarios) in the format that your system needs (CSV, JSON, API payloads), using the same notes and layering on any actual data that they have provided (screenshots, screenshare, mocked, or exports).

Example prompt

Now you're a Solutions Engineer who is translating a client's business operations data model into [Company Solution].
I have attached the notes and exports. It needs to be in this format.
Company: Company | Date | Price | Status
Project: Project name | Company | Due date | Assignee
Billing: Record number | Company | Sent date | Received date
Each object above (company, project, billing) needs to be its own CSV file. You can put them in prospect-data/acme-corp/.

4. Upload into the demo instance. A few options:

API: Connect to your solution's backend directly via the terminal.
MCP: Use a Model Context Protocol user testing tool like Playwright MCP or Chrome DevTools MCP to mimic user interaction via the UI. This can be done by bulk uploading from the admin backend or by adding new entries through the front end.
Manual: When all else fails, you can also just manually upload the data.

Let's assume that we can only upload how a normal end user can through the front end (no backend admin or API for some reason), using Playwright MCP.

Example prompt

Take the Object file in ABC Folder and start importing the data at yoursolution.com/object_name

The coding agent will first write the "test" script that automates these actions. Then it will run the test script (you can manually do this too in the terminal) npx playwright test [testScriptName].tsx

I've seen this process go from half-day, "block the afternoon" custom trial builds to 20-minute spin ups after the call. While the data uploading MCP is the biggest "wow" factor, the biggest time unlock is actually using AI to synthesize realistic demo data from a single discovery call, so you're not waiting on the prospect for their real data or burning creative energy inventing scenarios.

Full branch

Sometimes, you want to demo how your software might look in a future state if they sign a contract, and your company dedicates product development / professional services resources to change the functionality of your solution. We can now get out of Figma prototypes and slide-based walkthroughs (slideware) and into the realm of real-ish.

You'll take your real software, Git branch to your local machine (or through a vibe coding/hosted platform like Vercel's V0), and then make the needed mods.

Example prompt

I'd like to develop a module that sits on top of the company object and integrates with XYZ API Service. The example workflows/data should be [Insert discovery notes here]. You can mock fake API calls (just note that in the prototype). Users should access this functionality through UI entrypoint, and you should develop this in folder: app>api>components.

This is good for large accounts where you might need to bend your software to their needs, shape the roadmap to fit their requirements, or develop custom modules/services in the platform. Most common with pre-product-market fit tools or software that's bundled with professional services. The added benefit is you can swap out real API calls with mock returns to fully "own" the microservices layer and reduce fail points for a demonstration.

John Forbes is the founder of Aero and a former sales engineer & product marketer at high-growth software companies. john@aero.ws