Drag & Drop

The episode delivers moderate insight density with practical frameworks around AI governance, the agentic development lifecycle, and production challenges specific to government. However, significant portions devolve into abstract philosophy (people-first software, human empathy) and Outsystems product marketing that dilute substance. The most novel insight - that most government AI agents never reach production and that token costs create a 'token apocalypse' - is valuable but underexplored.

“the majority of them just don't get to production that they're in this proof of, this never ending proof of uh, concept state”

“agentic development lifecycle's much, well, uh, not much different, but it has that additional layer of tuning that happens after the fact”

The guest recycles standard frameworks (NIST AI Risk Management Framework, enterprise architecture thinking, human-in-the-loop governance) without meaningful new angles. The distinction between one-model-fits-all versus multi-agent/multi-model approaches is practical but not novel. The conversation largely confirms existing industry wisdom rather than challenging assumptions or offering counterintuitive takes.

“using that enterprise framework and uh, uh, a framework that's unique to our situation, our institution, our agency”

“one model to rule them all. It's not as simple as that”

Scott Ditch is a Principal Enterprise Architect at Outsystems with 20+ years in tech and documented work in government AI adoption. He demonstrates real operational experience across multiple agencies and brings practitioner credibility. However, his primary affiliation is vendor-side (Outsystems), which creates inherent bias, and no specifics about his direct role in shipping production AI agents are provided, limiting pure operator credibility.

“I'm principal enterprise architect for Outsystems”

“Been working in the tech industry since, what, 2003”

While the episode contains concrete examples (30-year-old payment system modernization, building permit workflows in California, tour booking use case), they lack quantifiable impact metrics, timelines, or financial outcomes. The building permit example is well-structured but anecdotal. Token costs and cost management are mentioned repeatedly but never anchored to specific pricing, volume thresholds, or actual budget impact data.

“customer that had a payment system that was uh, 30 years old, running on some pretty old technology”

“consider a building permit, you know, for counties and cities and stuff, that usually has quite a bit of labor”

The host (Faraz) asks generally competent setup questions but rarely pushes back or probes deeper. When Scott makes broad claims (most agents don't reach production, token apocalypse), the host accepts them without requesting evidence or asking how widespread these issues actually are. There are no moments of productive disagreement or uncomfortable follow-ups that would test the guest's claims. The conversation feels collaborative rather than investigative.

“Yeah, yeah. And you know, many, many of these agencies, obviously they have dozens of pilots, you know, going on”

“Yeah, and you know, with technology, you know, evolving”