A personal recap of GopherCon UK 2026 covering multiple talks: staying passionate as a developer in the AI era, AI-assisted coding case studies at Paddle, building Go web apps with HTMX/Templ/sqlc, treating MCP as the new API layer at Typeform, teaching CompSci with Go, zero-code Go instrumentation via OpenTelemetry/OBI, API design as a performance lever from Dave Cheney, building custom Go refactoring tools, and a follow-up London Gophers meetup with talks on goroutine lifecycle management and the Fyne-based FyshOS desktop environment. The author also offers feedback on talk length and speaker diversity for future editions.
Table of contents
How to Stay Passionate in the Age of AIKronk: Hardware Accelerated Local InferenceFrom AI Sceptic to AI Sceptic Who ShipsBuilding a Web App with Go & HTMXMCP is the new REST: making MCP Typeform's new APIHow Much CompSci Can Go Teach?Instrument Go Without Changing a Single LineAPI Design as a Performance PrimitiveBuild Your Own Tools to Refactor Go SafelyOne Definition of ValidBuilding a polyglot multi-agent AI architecture on AWS AgentCoreGo, ONNX & AI in Production: Trading PlatformBreadth of contentLondon GophersQuestions this post answers
How can MCP servers be designed so LLMs use them more effectively than a wrapped REST API?
LLMs work better with MCP tools that are semantically meaningful rather than autogenerated wrappers around a RESTful API. Typeform's approach, as shared by Andy Kuszyk, was to tune input/output types for LLM friendliness and move orchestration logic into the MCP server itself, and to build the MCP server directly into the web API so it shares the same business logic while not being constrained by the REST API's surface. Teams weighing REST versus MCP API design can track these architecture debates on daily.dev.
What is Dave Cheney's github.com/pkg/group library used for in Go?
It is a library for starting and stopping goroutines safely, addressing goroutine lifecycle and cleanup as a resource-management concern similar to how RAM must be treated as finite despite garbage collection. It distills years of prior approaches into a single solution for coordinating goroutine cancellation and cleanup in concurrent Go programs. Go developers managing goroutine lifecycles can follow concurrency pattern discussions on daily.dev.
How did Paddle use Behaviour-Driven Development with Cucumber to make AI-driven refactoring safer?
Paddle blocked their AI coding agents from modifying BDD tests written with Cucumber, which let large internal refactors proceed without altering user-facing functionality even though such refactors typically touch unit tests. This gave the team a stable test harness that kept AI-assisted changes constrained to internals rather than behavior, as described by Tom Coupland at GopherCon UK 2026. Teams building guardrails around AI coding agents can follow similar harness strategies on daily.dev.