OpenAI has published a customer case study showing how fleet management startup Proaction used the coding agent Codex alongside the voice and agentic models GPT‑Live‑1 and GPT‑6 Astra to lift sales conversion by up to 60% and save more than 75 hours of work per month. The standout detail: a non-technical co-founder now builds customer demos entirely on his own.

A Non-Engineer Builds Custom Demos with Codex

Proaction builds software for businesses that manage fleets of vehicles, from cars and trucks to construction machinery[1]. Personalizing a demo for each prospect used to require engineering time the small team could not spare, so the founders often fell back on conversations and slide decks instead[1].

Co-founder and COO Colin Knudsen says that by feeding Codex his sales-call recordings, prospect email threads, and shared spreadsheets, he can now generate an interactive HTML demo environment tailored to a prospect's own vehicles and workflows without looping in an engineer[1]. He builds four to six of these demos a month, each taking 30 to 45 minutes, which he estimates saves the team 40 to 60 hours of engineering effort monthly[1].

Knudsen says showing prospects a demo built around their own data has increased the share of deals that move from initial contact into solution development by 50% to 60%[1]. Beyond sales, he uses Codex together with tools such as Slack, Gmail, Linear, GitHub, and HubSpot to pull call transcripts, draft follow-ups, create issues, and update deal records — work he says now saves him another 25 to 33 hours a month[1].

Agents Take on Fleet Maintenance Through a "Managed Execution Layer"

Proaction also puts different OpenAI models to work across its platform. When customers submit photos of vehicle damage, ChatGPT‑5.6 Sol helps assess the issue, while the voice-capable GPT‑Live‑1 powers agents built for broader execution tasks[1]. The company calls this approach its "Managed Execution Layer" — moving beyond simply tracking fleet work to having agents actually carry it out[1].

One such agent, Marty, is designed to coordinate vehicle maintenance: it talks with a driver about a problem, calls repair shops, arranges service, and helps get estimates approved and paid, with Proaction staff stepping in whenever human review is needed[1].

Head of Product Danny O'Halloran says GPT‑6 Astra, the model OpenAI introduced in September, produces more succinct computer-use runs than the previous ChatGPT‑5.6 Sol did on comparable tasks[1][2]. GPT‑6 Astra is a major update OpenAI rolled out across both ChatGPT and Codex in September 2026[2].

Summary

Proaction's story illustrates how a small team is pushing AI use beyond sales support and into execution: a non-engineer builds demos with Codex, while agents running on GPT‑Live‑1 and GPT‑6 Astra now handle parts of the actual fleet-maintenance workflow. OpenAI is likely to keep highlighting customer stories like this to demonstrate what Codex and its agentic models can do across both development and day-to-day operations.

Source[1]: https://openai.com/index/proaction

Source[2]: https://openai.com/index/gpt-6-astra/