# [Call for Participation]: 2.5 hour Haskell Foundation Online Workshop > Source: > Published: 2026-06-24 03:00:23+00:00 Hello everyone, Please join us this Friday for 5 talks on the uses of AI in Haskell development. The event will stream on our [YouTube Channel](https://www.youtube.com/@HaskellFoundation) starting at 11:30am Eastern on Friday, June 26th. And will go until 14:00 Eastern. Here’s the scheduled program (talk abstracts are below) 11:30 Alfredo Di Napoli: Consulting for clients who use AI: a tale of a few hats 12:00 Gautier Di Folco: LLM as a SDLC support 12:30 Christian Georgii: Tricorder, A new development tool for Haskell and LLMs 13:00 Lorenzo Tabacchini: Though this be madness, yet there is method in’t 13:30 Nathan Bloomfield: Typed Holes + LLM + GHC Plugin It’s a good mix of tools and experience reports, I hope to see you there! As individuals, we are in charge of our own development processes and tools; as employees of a company, we must adjust them to fit the company’s best practices and blessed workflows; as consultants, we must do all of the above and comply with the client’s own processes – and if those include substantial use of LLMs, how do we navigate all of that? The last part is a challenge in itself: how to balance a consultancy’s *ethos* of producing the most *valuable* output from an hourly rate with the temptation of offloading some of the mechanical work to an AI? What does “most valuable” even mean when machine-assisted development is in the mix? In this talk I’ll try to give my own answers to those questions by presenting my personal and professional experience using LLMs to design, write, test and review code, ranging from recreational projects like [Rusholme](https://github.com/adinapoli/rusholme), where human-supplied domain expertise drove the design of a toy Haskell compiler, to my daily job as a Well-Typed consultant working with clients who make heavy use of LLMs and encourage us to do the same. In this experience report I’ll show what worked for me and what didn’t – and that ultimately it comes down to wearing a few hats: most of the time an “architect”, sometimes a “coder”, and, occasionally, knowing when to take the hat off entirely. Building a highly secure communication toolset requires rigorous engineering, especially when maintaining a robust Haskell backend. In this session, we will explore the practical realities of integrating AI into a daily workflow. Moving beyond the hype, this talk dives into the actual development trenches to demonstrate how to effectively leverage LLMs for complex backend engineering. We will cover actionable techniques for optimizing context windows to yield highly accurate results, utilizing AI to draft compliant Pull Requests, and streamlining the debugging process when things go wrong. Finally, I will share candid insights from my hands-on journey testing and evaluating multiple models (Claude, Gemini, GLM, and Gemma) highlighting how each handles the unique challenges of Haskell development and tight feedback loops. [tricorder](https://discourse.haskell.org/t/ann-tricorder-a-new-development-tool-for-haskell-and-llms/14208) has been my daily driver for Haskell development for months now. It works really well as a standalone replacement for ghcid or ghciwatch with some extra features that make it very attractive: However, it really shines when pair programming with an LLM agent. The included CLI surfaces build information in a way that is context-aware and reliable, so the feedback loop is very tight I wrote a compiler for Constraint Handling Rules, almost entirely through agentic coding. In this talk I will show you how it went, why this compiler seems to be an exceptionally AI-friendly problem, and what I learnt about agentic coding in general, with a handful of suggestions you can test against your own work. Typed holes are a handy design and reasoning tool available in languages with sufficiently strong type systems, including Haskell. They stand in for code that isn’t yet written but which the typechecker can still infer useful information about from context available at compile time. OllamaHoles is a GHC plugin by Matthias Gissurarson that uses an LLM to generate valid hole fit suggestions. This talk tours how the plugin gets the LLM and GHC to cooperate by doing what each one does best, with some reflections on the presenter’s experience using an LLM as a development tool.