Optimizely London Developer Meetup 2026: my recap

On 2nd July 2026 the Optimizely London Developer Meetup returned to The Lightwell. Six talks, six practitioners, and one running theme: the interesting engineering lives at the seams, where SaaS stops mirroring PaaS, where Coveo stops being Find, and where edge orchestration ends and your .NET tools begin. Here is my account, talk by talk, with the architecture decisions I would flag if I were scoping any of this for a client. 

Scott Reed (Niteco): Replacing Search & Navigation with Coveo 

Scott walked through a live project replacing Optimizely Search & Navigation on a Customized Commerce build with Coveo. The honest headline: Coveo is a platform shift, not a drop-in Find replacement. Features like boosting, ranking and pinning do not map one to one, they become different Coveo workflows. Two lessons stood out. First, push and stream your catalogue via scheduled jobs rather than crawling, so you own freshness and structure. Second, the SEO pivot: a pure client-side Coveo storefront is invisible to crawlers, so they moved to hybrid SSR. If you are evaluating headless Coveo, lock that decision in before the POC, not after. And treat analytics validation as a first-class deliverable: no analytics, no AI. 

Guilherme Menezes (Cloudflare): Agent. Sandbox. MCP. 

Guilherme built one agent from Cloudflare primitives in roughly a hundred lines of TypeScript, with MCP as the door into the CMS. The interesting bit for our world: that TypeScript is only the orchestration layer, the part you own does not move. The MCP server exposing your CMS tools can happily be .NET sitting alongside Optimizely, the same shape as Opal and Optimizely.Opal.Tools. Two things I would flag as a TA. Identity has to be real end to end, so your MCP tools must act as the calling user and honour roles and content access rights, not run as a service principal holding every key. And it is not Opal-or-edge; they meet at the same tool surface, and the real decision is where you draw the boundary. 

Minesh Shah (Netcel): SaaS CMS UI Extensions 

Minesh tackled the gap most of us have felt: PaaS gave us the keys to the building, SaaS does not. UI Extensions close that gap without handing a divergent CMS fork back to you, the framing being "same codebase, many editor experiences." There are three plug-in points: sidebar apps, full-page apps, and property editors (his flight-fares editor resolving a product ID to live pricing was the standout). It is all built on OCP, so if you have shipped OCP apps the model already fits your hands. Best of all, it is safe by design: sandboxed iframes, scoped permissions, and versioned host APIs. Pin a version and stay stable as the CMS moves underneath you. 

My talk: CMS 13 & Commerce Connect 15 

I spoke on CMS 13 and Commerce Connect 15: what has changed and what to watch for once you are shipping. Three shifts matter. .NET 10 is mandatory, LTS and only .NET 10, so a single package without a .NET 10 build blocks your whole restore; audit before you scope. Optimizely Graph is now foundational in every licence, and Search & Navigation is removed, not deprecated, so search, filtering and product search all need re-platforming (and mind the Graph type hierarchy, a base type query will not return derived types). Visual Builder is the default surface with on-page edit off by default, which editors feel on day one. Commerce Connect 15 rides the same .NET 10 foundation, so upgrade the two together and treat it as a coordinated project, not a patch. 

Simon Chapman (Optimizely): the 2026 roadmap 

Simon walked us through the roadmap under three themes: an Opal-powered agentic CMS, CMS 13 as the modern PaaS foundation, and continued core investment. The 30th June Opal drop was substantial: a 45-plus agent library, agent sharing and reporting, an AI image editor, and multi-model support so admins can pick Gemini or Claude, org-wide or per agent. Underneath sits the Agent Orchestration Platform: evals and guardrails, no-code workflow orchestration, and a Graph, Analytics, RAG and Memory context layer, with interop via Custom Tools, Webhooks, Remote MCPs and A2A. The governance piece I have been waiting for, Visual Builder content-model locking, is on the way. 

Andrew Markham (UNRVLD): SaaS, reevaluate your approach 

Andrew closed the evening with a pragmatic message: moving to SaaS CMS is not a like-for-like port of your PaaS habits, some extension points simply are not there. Code-first content modelling still works, but he was candid about the gaps: no code-set default values, no custom property types or selection factories, and no custom validation rules. Rather than treat those as blockers, his operating model is a sound one: audit your content model against the limitations upfront, design within the constraints instead of fighting them, and feed the gaps back to Optimizely as the platform matures. 

Main takeaway 

If I take a single decision back to client work, it is this: no fixed scope until we have run a proper platform-fit audit. Every talk pointed the same way. Coveo's SEO pivot, the .NET 10 restore trap, the Graph type hierarchy, the SaaS content-model gaps, identity in the MCP layer; none of these are build problems, they are discovery problems that only bite late if you let them. So I would move that audit to the very front, before a line of code and before anyone signs off a number. The seams are where the risk lives, which is exactly where the discovery effort should go too.