AI Automation Dojo by Office Samurai

There are genuine operational insights scattered through the episode - using Power BI as a security buffer between the agent and SAP, limiting responses to 10 invoices per vendor, excluding amounts and cost centres due to 'toxic combinations', and the idea of MIA serving as a backend engine for other automations. But these are buried under long stretches of vague AI-is-changing-fast platitudes, meandering tangents about Copilot in Excel, and the host's personal anecdote about Python and Gemini. The density is moderate at best.

“actually the data source directly, uh, from the agent is actually Power bi, like basically data model from Power bi, uh, that we have used. We didn't go directly to SAP”

“if you think about like 50, 60% of the queries requires in the first step to check what's the invoice status”

The reframe of shifting from a ticketing paradigm to live chat as a fundamentally different service model is a useful observation, and the idea of MIA acting as a callable backend for other automations shows some first-principles thinking. However, the majority of the episode covers well-worn territory: AI governance concerns, not over-scoping v1, the RPA-to-agent evolution, and 'AI is moving so fast' hedging appear throughout without fresh angles.

“the idea is behind, we hope that actually MIA would be playing a key kind of background role. So you know we don't need to build an extra, extra database uh, uh to get like invoice statuses”

“we decided that we need to make some cleanup and stabilization first like uh, from system to process also to take that into account”

Matt is a genuine practitioner who has shipped 200+ automations and just put a production AI agent into a SOX-regulated global finance environment - he is not a thought-leader or career podcast guest. His domain specificity around P2P edge cases and data governance decisions is credible. The ceiling is held down by Senior Manager rather than C-suite seniority and by the fact that the host is also his vendor, making the conversation partly a joint case study presentation rather than an independent interview.

“I'm responsible for the team of uh, Intelligent Process Automation, which is like uh, if you recall, old rebranded, uh, let's say Robotic Process Automation team”

“P2P first of all was really heavily involved in that project and they were really, really supportive”

The episode offers a handful of concrete details - 30-minute Power BI refresh cadence, 10-invoice cap per query, 14,000 employees, 50 - 60% of P2P queries needing invoice status first, SOX regulatory context - which is more than most AI podcast case studies provide. What is conspicuously absent is any outcome data: ticket volume reductions, hours saved, time-to-response benchmarks before and after, or cost figures. The agent was freshly launched at recording, which is a legitimate reason, but the episode would score substantially higher with even early-signal metrics.

“Power BI is refreshed once in a half hour”

“having in mind that we have 14k employees. We would not like to have role level security by the persons”

The host has genuine domain knowledge and lands a few good questions - probing the amounts-removal decision, the jailbreaking attempts, and the scope rationale - but the inherent vendor-client dynamic (Office Samurai built the tool) keeps the conversation promotional and unchallenging. The host frequently answers their own questions, inserts long personal anecdotes (the Gemini/Python loop, the mortgage bank story), and never pushes back on any claim or asks for hard evidence of impact.

“did you on this um, kind of uh, breaking or jailbreaking, uh, did you catch anyone trying to jailbreak your meal?”

“I think it's worth noting for the people who are thinking about such a project that uh, this data security aspect is extremely important”