The most expensive AI is the AI you deploy wrong
Every company can buy AI today.
Few get value from it.
Getting value from AI comes down to one question: Which kind actually fits the problem in front of you?
In this webinar, we will look at this problem from two ends. David Bečvařík will focus on AI that acts inside a real business, and what has to sit underneath it before it can carry production. Outages, cost limits, audit, human approval. Without that, an agent is just an expensive experiment. Antonín Šimek will present a case where AI runs on the device itself. No server, no data leaving the phone, no cost per request. He built it into our own event app, and he is clear about where it stops.
The session is intended for digital, product, architecture and security leaders in banking and finance, energy and utilities, and retail and e-commerce.
The session is held in English.
How much AI a problem actually needs
AI comes in very different sizes. A deterministic system. A small model on the device. A simple cloud call. A full agent with production infrastructure behind it. Each fits a different kind of problem, at a very different cost. Companies often default to the largest option and use it for everything. That is where budgets overrun, output cannot be trusted, and compliance is put at risk.
AI that takes action, and what it costs to run safely
When an agent only answers, a wrong reply is easy to fix. When it acts, it can do real work on its own, and a wrong step can move money or change a record. What keeps it safe is everything around the model: a stop on risky actions, a person to approve what matters, limits on spend and steps, and a log of what it did. None of that shows in a demo. It is the whole job in production.
The use of on-device AI
When a local model runs on the user's own device, there's no network call, no latency, no data leaving the phone, no cost per request. For a bank, an insurer or a retailer, that turns privacy into a built-in property and takes the cost of scale off the table. We put this into practice using on-device AI with Apple Foundation Models. We are also honest about the ceiling, the point where on-device stops being enough and you fall back to the cloud.
Questions you will be able to answer
Does this problem even need AI, or would a deterministic system do the job cheaper and more reliably? When is a full autonomous agent the wrong call? What has to be in place before an AI system touches production in a regulated environment? Where does a small on-device model beat an expensive cloud one?
Agenda
David Bečvařík presents
- What has to sit under an AI agent before it can carry production
- One real client project: where AI created value, and where control made the difference
- Surviving outages, guardrails, human approval, cost and audit limits
- When not to reach for a full agent
Antonín Šimek presents
- On-device AI: no server, no latency, no data leaving the phone
- A concrete example from a real app, and where on-device stops being enough
Joint Q&A
- Ask our speakers
