Proaction Codex: How a Fleet Software Startup Turned Sales Demos Into a Solo Engineering Task

Proaction, a company that builds fleet management software for vehicles and construction machinery, has changed who gets to build a sales demo. With Codex from OpenAI, co-founder Colin Knudsen now creates customized interactive demos himself, without pulling an engineer off other work.

One person, four to six demos, no ticket queue

Before Codex, a customized demo for a prospect meant asking an engineer to build an HTML environment that reflected that prospect's actual vehicles and workflows. Now Colin Knudsen does it alone. He feeds Codex call recordings, prospect emails, and shared spreadsheets, and Codex uses that material to customize the demo environment so it shows the prospect's own fleet rather than a generic mockup. Each one takes him 30 to 45 minutes to build, and he produces four to six per month. Proaction estimates that comparable demos would take its engineers about 10 hours each to produce, which puts the monthly engineering time avoided at 40 to 60 hours. For Knudsen personally, Codex is saving an estimated 25 to 33 hours a month.

What a personalized demo does to a sales conversation

The reason this matters beyond time savings is what a prospect sees when they open the demo: their own vehicles, their own workflows, rendered inside the tool they're being asked to buy. Proaction estimates that these customized demos increased the share of deals moving from initial contact to solution development by 50% to 60%. That's a claim about pipeline movement, not about revenue outcomes, but it's a meaningful signal that specificity at the top of the funnel changes how prospects engage. Proaction extended the same logic into a customer solution center, also built with Codex, where prospects can explore tailored workflows and review sales materials in one place rather than piecing together information from a sales rep.

Codex wired into the tools the team already uses

None of this happens in isolation from the rest of Proaction's workflow. Codex plugins connect the company's processes with Granola, Gmail, Slack, Linear, GitHub, and HubSpot, which means the demo-building and solution-center work draws on the same systems where calls get recorded, deals get tracked, and code gets shipped. That connective layer is what lets one non-engineer move from a call recording to a working, customized demo in under an hour.

The same model family, now inside the vehicles themselves

Codex is only one part of how Proaction uses OpenAI's models. On the product side, the company uses ChatGPT-5.6 Sol to help identify damage in vehicle issue photos, and it's building fleet-operation agents with GPT-Live-1 under a system it calls the Managed Execution Layer. Those agents are built to handle tolls and vehicle service workflows, and their range of actions includes making voice calls, reviewing documents and images, analyzing text, and responding in chat. One named agent, Marty, coordinates vehicle maintenance: it can discuss a vehicle problem with a driver, contact a repair shop, arrange service, and help get an estimate approved and paid. Proaction still applies human review or intervention where agent work requires it. GPT-6 Astra, meanwhile, is being used to build these agent experiences faster, which suggests the sales-side pattern of one person doing what used to require a team is being tested against a harder problem: running physical fleets, not just selling software to run them.

Specs

• Demo creation frequency: 4 to 6 per month
• Demo build time: 30 to 45 minutes each
• Estimated engineering time avoided: 40 to 60 hours per month
• Estimated monthly time saved for Colin Knudsen: 25 to 33 hours
• Models mentioned: ChatGPT-5.6 Sol; GPT-Live-1; GPT-6 Astra

Sources

• OpenAI — OpenAI