Dan Aberhart opens a GTF Productions live session on a farm’s busy-season paperwork and the question he hears most: what can I take off my plate first? The session then follows the room. Audience members ask about tracking grain basis behind a company login and reconciling soil test lab reports. Dan builds a basis tracker demo, a soil test tracker and a personal morning dashboard by voice while they watch. He is open that the basis numbers in the demo were fictional, and that on data security he is a visionary, not a security expert. The takeaway for people in agriculture is that the starting point is context and one small job, not a tool.
Five invoices a day. That is what a CFO friend of Dan’s says a farm can see in a busy season, and each one has to be checked, categorized and matched to a receipt. Most people in agriculture have nobody to hand that to, which leads to the question Dan hears most: what can I take off my plate first?
This live session starts there and then follows the room. Dan sets out to show one job and ends up building three things in front of the audience, answering a hard question about security, and being candid about what is still a demo. He is not a machine learning expert and says so. What he does offer is a working method.
What’s Inside
• Why the first step is context, not a tool, and why Dan says dictating to the machine beats typing to it
• A grain basis tracker for a terminal that keeps its prices behind a password: ask the company for a supported connection first, then look at what browser and computer control can do
• Why the basis numbers in the live demo were made up, and what a verified source for one specific terminal still has to prove
• A soil test tracker built live from a few questions: matching lab shipment receipts to returned reports, then checking what the grower portal shows
• A morning dashboard changed by voice while the room watched: weather, email summary, market news, futures report
• Where to start with agents: one agent that does one job, then a small group, named and organized like staff
• The privacy question, answered without a sales pitch: what goes to the cloud, what can stay local, and why computer control is the feature to be most careful with
• The tire torque reminder: an assistant that read an invoice email and flagged a safety item nobody asked it to watch for
Dan opens with a problem every farm office knows: invoices, receipts and the people chasing them in a busy season. He says the real question people bring him is what to take off their plate first. He tells how he rebuilt his own business with a team of AI agents. He also tells how an assistant read a tire shop’s invoice email and reminded him to get his wheel nuts re-torqued, something he had not asked it to watch for.
Dan had planned to demonstrate one job, an invoice dashboard, but the audience questions take the session elsewhere. One participant wants to track grain basis over time from a grain company’s login-protected site. Dan explains the options in order: ask the company whether a supported connection exists, then use browser or computer control with a login. He also sends his voice agent out to research verified sources. It finds basis history for one regional elevator on a commercial data service. It cannot confirm coverage of the participant’s exact terminal or the licensing. The dashboard Dan builds for the demo uses fictional numbers, and the participant points that out.
A second participant wants to reconcile soil test shipment receipts with returned lab reports and check that a grower portal was updated. Dan has his voice agent interview her for the key facts: the matching identifier, the portal and the volume. He then builds a soil test tracker demo with fictional data. The demo screen warns that a field ID is a clue and not a unique key, and that a received report does not prove the portal was updated.
The session closes on a direct question about security. Dan answers as a user, not an expert. Most people, including his cohort, use cloud models. Computer control is one of the more dangerous features. Sensitive intellectual property belongs in local AI. Everyone should do their own research. He then describes his upcoming AI training cohort and answers questions about format.
Key Topics
• Starting with one small job and the context you give the machine, not with a tool
• Dictation as the fastest way to give AI context
• Getting at data behind a login: supported connections first, computer control second
• Matching lab shipment receipts to returned soil test reports, and why a received report does not prove the portal was updated
• Data security and local AI, from a user’s point of view
Resources Mentioned
• Claude Code, including its Chrome extension and computer use feature
• OpenAI Codex, which Dan prefers for voice
• Wispr Flow, for dictation (spelled “Whisper Flow” in the captions)
• A commercial market data service and a grain company’s basis page
• Connections Dan names: Asana, Canva, Descript, DocuSign, Gmail, Google Drive, HubSpot, Mailchimp, Notion, Zoom, QuickBooks
Connect with Growing the Future
Website: growingthefuture.ca
YouTube: Growing the Future
Instagram: @growingthefuturepodcast
LinkedIn: Growing the Future
Subscribe to the Growing the Future Podcast wherever you listen, and register for Convergence Conference 2027 at convergence.ag.