This is a place for practical writing — not press releases, not product announcements, not vendor talking points. Just observations from inside enterprise AI rollouts, especially in fintech and regulated industries.
What you'll find here
A few patterns we keep coming back to:
- The shape of AI failure. Most AI projects don't fail at the demo. They fail at the second meeting with security, the third invoice from the LLM provider, or the moment someone asks "is this auditable?"
- What production really looks like. The difference between a Notebook prototype and a system 500+ people rely on isn't engineering effort. It's design choices made on day one that compound for years.
- Vendor-agnostic thinking. Most consultancies are partner programs in disguise. Here, the question is always "what does your business actually need?" — not "what comes with the highest referral fee?"
- Frameworks over advice. Specific decision trees, reference architectures, and scoring tools. Things you can copy and use, not slides to skim.
Cadence
New writing every two weeks or so. Substack-style depth, but on our own domain — so the work compounds in one place, indexed by search engines, owned by us and shared on our terms.
What's coming
The first few pieces will tackle:
- A reference architecture for production AI — the components, the runtime, the observability layer, the security model. With opinions about what to skip and what to never skip.
- The silent killers of AI agents in production — 12 specific failure modes seen across enterprise rollouts, with concrete fixes.
- Build vs. partner vs. buy — a decision framework calibrated for regulated industries, where the trade-offs are different than in consumer tech.
- AI cost surprise — the hidden lines in your AI invoice, and the architecture choices that prevent them.
If you want this to land in your inbox or have something specific you'd like to see covered, start a conversation.
—
RaptorB Consulting