On Oct 29, 2026, at 6:21 AM, Winfield, Alister (Senior Solutions Architect) via Freeradius-Users <freeradius-users@lists.freeradius.org> wrote:
Second that scaling RADIUS is hard...
And for what it's worth, AI doesn't help a lot. Ignoring any AI mistakes or hallucinations, AI largely does what you ask it to do. And, it takes shortcuts where it can. "Vibe code a RADIUS server compatible with RFC 2865 etc." will get you something for maybe $50 in AI tokens. However... * the RFCs are often wrong * many, many, things in the RFCs are left implied, and are not documented * the RFCs don't describe interoperation problems. Vendors have bugs, and the RADIUS server needs to work around them * this isn't just RADIUS, it gets worse when you add EAP, TLS, multiple EAP types, etc. And that's just for writing the software. What about running a RADIUS server? A vibe coded server will have issues: * no logging unless you know what you want to log * no monitoring unless you know what you want to monitor * no debugging, AI will do that for you! * messages will be writing in something resembling English, but will be vague and not actionable On top of that, the system, won't address any of the issues Alister brings up. And even if you succeed in solving all of those issues, congratulations! You now have a vibe-coded product that no one else in the world uses, and that you have to maintain yourself. :( Hence (a small plug) my webinar later today: https://www.inkbridgenetworks.com/videos-webinars/inkquire-tools-for-freerad... It's taken me and my team 3 months to come up with that suite of tools. I suspect that it will take anyone else more time than that. Back to the original question, it's simply impossible to create a "canned" product which does what everyone wants, and which scales. Even in the age of AI, the problems are just too hard. Alan DeKok.