Hi Max, if I understand this alongside some of the other examples you've posted, is it the case that this new side bar which references skills replaces the need for the granular, single purpose formulas structured in tables that you demoed in detail with the insurance use case? In other words, is this an easier way to build the same functionality that you would previously have done with cascading AI formulas? Or has this opened up a new capability for fuzzier, less deterministic use cases like this that require checks and iterative loops?

@Max O'Brien thank you and looking forward to the next instalment!

Thanks Max and great demo again. Half way through I was wanting to ask whether you've explored the OpenAI or Gemini packs and why you prefer the native AI. (My concern was lack of control over the model, being worried about future pricing etc.) I now realise that my decision to have used the OpenAI pack for my own testing has been a huge mistake. The issue is that automations don't know when the OpenAI formula has returned a response in the way the native formulas do, unless you force them inside a series of nested withNames. As a result you must then bake all your logic into a single huge action, or have timed automations cleaning up downstream if you can no longer handle the complexity. So lesson learned on my side. For context, I implemented something a little less complex than what I think Max did here using the OpenAI pack, parsing a dictated note and sending it to one of several tables, but this is what my formula looks like (with more downstream too): RunActions( /* STEP 1: Determine the Table (Synchronously) */ WithName( OpenAI::ChatCompletion( [Michael Dean (1)], thisRow.[User prompt], "Interpret the user’s intent and decide which of the following tables they want to interact with. Respond only with the tableName from:" + [Convert Table to CSV or JSON]::tojson( SchemaTables.TableName, "tableName", SchemaTables.Description, "description" ), "gpt-5-nano" ), PredictedTable, /* STEP 2: Determine the Operation (Synchronously) */ WithName( OpenAI::ChatCompletion( [Michael Dean (1)], thisRow.[User prompt], "Interpret the user’s intent and decide whether they intend to make a create, read, edit or delete change to an entry in a table containing Tasks, Ideas, and Daily Logs (health, sleep, social). Step 1: Synonyms and usage: Create: Add, new. Also includes directives to capture/process content (e.g. 'Summarise this idea', 'Log that I slept well', 'Remind Tali to cancel'). Read: Show, list, display, return, what are, tell me. But NOT(ask, review, check).

Hi Max, thanks for kicking this off. I'm Michael, based in Australia and working in a big corporate in a finance automation capacity. I support a team of regular finance analysts who need to maintain and extend anything we build, so your framework of cascading formulas in an Excel-like environment resonates. I've been implementing a version of what you demoed in your NoCode agents video in my personal life after having failed in Make. It's going great and I'm keen for the sessions to kick off. Thanks again for sharing!