The collision with Nathanael's changefabric.com: different name space. His platform runs on
API and cloud token spend. This one runs on local Claude Code subscriptions, on developer
machines.
Proof of the same pattern
Not a separate story - the same delivery pattern, twice more.
CAS 360 - "chat with your data" for Come and See Foundation, built solo in weeks, in production, handed off fully
customer360-example - the same build abstracted into reusable sales collateral
Leo's open questions, with a point of view grounded in what's shipped.
Lead: standardize infrastructure on Docker
AMFM and customer360-example are already Docker-first, terraform-managed
Docker is the common interface nearly every host supports
Local development actually mirrors production
FaithBase, where I hit the most friction
Vercel itself was fine for me. Convex was the real friction - a major divergence from every
other platform choice I'd made, and real effort to get comfortable with. Not a verdict on
Convex: it's opinionated in ways some developers want and I wasn't used to.
AWS, where I've hit the least
My experience, not an objective ranking - someone else with no AWS guardrails could hit more friction there than on a more opinionated platform
For me: AWS can build anything, and it's the core primitive underneath the other platforms
Same argument applies to Google Cloud or Azure; AWS is my personal preference
Recommendation: more direct AWS work, simplified billing back to the client
On-prem compute
Most of the team works off MacBooks. When the lid closes, AI compute stops.
Cloud environments don't have local credentials or filesystem access, and cost more than compute already owned
Dedicated on-prem hardware per developer, or shared in the Franklin office