The future of programming, using github copilot workspace and pythagora-AI

Coding with agents

To summarize my experience after playing around with copilot workspace and pythagora-ai is that the future is not here yet. Or maybe this is what it feels to be a project manager. You give up control and watch the AI mininterpret, hallucinate and remove or break old functions when adding new features. However I still must say that it is getting better, and it is a fun experience.

Here is the article where I tested pythagora-ai.

Creating a simple TODO App with Pythagora V1 (Early access)

If you want to learn how to use the copilot workspace, look here:

https://github.com/githubnext/copilot-workspace-user-manual

As a starting point I used claude-AI sonnet 3.5 (New) to get a useful first version. This is in my opinion the best AI coder that I have tested so far. https://www.anthropic.com/claude/sonnet

After that I gave the following instructions for a change request in github workspaces.

Add another page as a new html file (frontend/users.html). 
It should allow editing the user data and use the appwriter database as a backend.
Also add a backend/create-database.js node script that connects to the database
and create the data. Here is an example of how to  fetch data from Appwrite in the frontend:
databases.listDocuments('EVENT_TIME_ID', 'MEMBERS_ID')
        .then(function (response) {
            let records = response.documents.map(function(doc) {
                return {
                    id: doc.$id,
                    name: doc.name,
                    phone: doc.phone,
                    replied: doc.replied,attending: doc.attending
                };
            });
            w2ui['grid'].add(records);
        })
        .catch(function (error) {
            console.log(error);
        });

After this step you can brainstorm a little,

Brainstorm

After the brainstorming step, you generate a plan,

generate a plan

After implementing the plan you can create a pull request and merge it.

After testing and merging the pull request, you can safely delete it.

We cannot fully trust the AI to generate fully functional apps yet, and there is no good way to let the AI test the app generated, but I think we will get there. Agents are here, and coding agents are really helpful to get you started with at technology you know nothing about. In my case this was appwriter

This is what the add ui looked like:

user interface, to add users

This was the user interface for the complete event

This was the inital idea for the application,

Create a web app uses appwriter and handles confirmations for an event.
There should be a deadline and event location and time, as well as members 
to the event.

When the future is here, we will no longer be code monkeys. We will be test monkeys improving the specifications until we get the desired result.