Tech Talk NOW

I’m a big fan of the current @united safety video…I like the Rube Goldberg ball rolling machine. Or maybe I have just seen it too many times in the last six months and have convinced myself it’s enjoyable…

Over the last 3 months, I've seen enterprises working with us on moves to Fusion, S/4, Dynamics, Infor, Sage, EPM, Workday, ServiceNow, Salesforce (CRM), Snowflake, Databricks, and (really!) consolidated AS/400 environments. I've seen source landscapes spanning dozens of custom apps, legacy SAP, PeopleSoft, JDE, Baan & Lawson, EBS, Hyperion, mainframes, and (again) AS/400s. My main takeaway is that no enterprise is a single-environment shop. Even a landscape which prefers one vendor over another has critical issues in boundary systems and consolidation risks, and every source/target has its own flavor. - Oracle - design is the critical factor - SAP - semantic differences in identically named sources combined with (seemingly) trivial breakages that end up becoming critical - Dynamics - explicit extension boundaries become interesting choices - Infor - even finding out the version/product you're on is a challenge… - Workday/SNOW/SF - SaaS may have simplified deployment stacks but it did not simplify business processes… - Snowflake/Databricks - the question of what do I actually want sometimes comes into the picture far too late… - AS/400 - it does work…and that's enough for a lot of enterprises. I don't think of AI in terms of tooling or operating systems for these landscapes. That gets into the territory of vendor-specific copilots when no enterprise is a single-vendor shop. I instead think of AI as digital labor to automate the traditionally heavy task of maintaining and transforming such systems…which is a very different proposition. It requires a higher degree of trust, autonomy, and even agentically driven validation processes. But that change is worthwhile - given that the ROI is substantially higher moving from a human-augmented-with-AI model to an AI-driven model with humans involved only at specific gating points.

The conversation around renting frontier intelligence versus hosting models almost feels like a mirror of the argument between on-premise data centers and hyperscalers…except that now hyperscalers have been “accepted” as part of the secure landscape and sovereign infrastructure of a customer. The fundamental difference I see is trust. Hyperscalers invested massively into security, outreach, and compliance. Meanwhile, all we hear from “AI-scalers” is stories of “uncontrollable agents breaching sandboxes with no controls”. The difference is stark… “Enterprise sovereignty” has a fluid boundary. If VPCs and hyperscalers count as part of sovereign infrastructure, then the boundary really shifts at the speed of trust. I cannot say that another lab AI breach every week inspires a great deal of trust.