Was looking around for examples of git for complex document/policy management outside of software/code and found this repo where the Council of the District of Columbia stores its statues and code as code in a git repo with a full history of commit changes, diffs etc.
It was amazing days in DC with Tom MacWright building the initial tools to make this so. It'd be so amazing for all governments to do this, to make the law (& how it changes over time) readily available.
Exactly - the amount of transparency provided by this commit history and diff is really cool. So many ways git features provide some interesting ideas e.g. could PRs be a way to track changes from different parties or even citizen groups? Could PR reviews/comments be a way to capture public discourse?
AI building the integrations is the approach we are taking at the moment, to the extent that you can store credentials like API keys and Oauth and then make the AI aware of these via a variable like ${googleAPI.SECRET}.
These credentials can then be use in the API node (where the AI can write a custom curl) or the run code component (where the AI can write custom python or js code to make an HTTP request).
Can it automatically update themselves based on changes to the source API? Can it detect ahead of time if APIs have changed? How about understanding the API requirements?
Interesting idea, I have it a go with this prompt: "Generate a workflow that uses code to scrape the content from a page, passes it to an AI assistant that then checks it for any types, and then outputs the typos and the correct spelling"
To even consider for enterprise M365 use, it needs to be 1:1 with Power automate/Flows/Apps, Entra, Intune, Graph API, PowerShell, AAD/Azure, etc. If I need to create a new user, it's going to involve every corner of M365/Azure. Maybe getting full functionality with Flows and PowerApps would be a starting point.
Our AI goes much further than Zapier/Make in terms of how far it can get you towards a complete, ready-to-run workflow.
Make's copilot is pretty limited to generating an outline of the flow by selecting the right nodes but does not actually configure them. You still need to manually click into each one and set it up.
Zapier goes a bit further than Make, but it still leaves the workflow with a lot of configuration work that needs to be picked up by the user.
In both Make and Zapier, you really need to prompt the AI copilot in a very specific way to get good results. In our case, the AI is designed to use its business analyst/consultant mode to extract information so it can work from very general, unclear and ambiguous instructions to a clearly defined workflow/process to build.
The ability for our AI to edit the workflow at any time (including on top of your own manual changes) also means you can have a continuous iterative dialog/interaction with our AI copilot vs a once off interaction at the start. Both Make and Zapier's AI Copilots lack this or are very weak in being able to edit existing workflows reliably.
What exactly is "your AI"? From how you describe it, it's a GPT model with a "You're a business consultant" prompt.
I'm sorry to be rough, but from your description it just sounds like your AI somehow does a better job with prompts compared to Zapier & Make, which is highly subjective.
Besides that, this looks very cool and IMO is the future of interfacing AI automations in work environments
Valid point. At its core, our AI is indeed a LLM that is prompted to provide an output. A lot of the work however is in (1) prompting it in a way that allows it to actually understand the user instructions and current state of the workflow and (2) allow it to reliably output a response that acts as a set of instructions on what needs to be done within the platform e.g. add this, remove this, change this question text to this, write this code etc.
One example of what we had to do to achieve this was to develop an "intermediary language" defines how the current state of the workflow is represented to the AI and how the AI responds back - this needed to capture enough detail about the workflow without overwhelming it with too much context. We also developed techniques for structuring the prompting, with the process of building a workflow actually split into 3 stages: a pre-build planning stage, a build stage where the overall structure of the workflow is set, and then a build node stage where each individual node its configured. There is a bunch of other techniques we developed to get LLMs to be able to do what they current do, but these are just some examples of how it's a bit more than just a "You're a business consultant" prompt.
One thing I'd encourage people to do is test these co-pilots head-to-head on the same prompt. If you were to ask Zapier or Make to "build me a process for triaging customer complaints", I'd expect them to not get very far, perhaps an outline of some apps you could connect together to achieve it. If you asked our AI this same request, it would be able to deliver a complete workflow with fully configured forms, tables, branching logic, tasks etc
Partially inspired by WD-40 but more so the exorbitant price for anything workflow.com. The number did have some logic behind it: 8 letters in automate and 6 in no-code.
