Jeff Hale

@cursor_ai I love the new features! But tell me, is it time to abandon Azure Devops for Github since it clearly looks like you're valuing the cloud agent approach. Please and thank you.

Cursor can now continuously monitor and improve your codebase. Automations run based on triggers and instructions you define. cursor.com/blog/automatio…

We're introducing Cursor Automations to build always-on agents.

I've been slowing automating development work to OpenClaw using the Opus 4.6 model and its amazing. New workflow unlocked: Raygun error reported -> Pull code and suggest fix -> Make change and create a PR. Slack messages keep me in the loop.

NotebookLM is amazing. I gave it my notes to a developer talk I presented to my team and it put together a video that is better than my presentation!

I've been playing with OpenClaw the last couple of weeks. An AI personal assistant that runs on your own hardware and can help automate anything you can think of. So, building upon the planning feature I have within a project story, I've created an API for it, where it can pick up stories with finalized plans, implement the code, create a PR, and then mark the story as processed. This instance of OpenClaw has a user account to Azure Devops. A user account to our project management system. And a bot configured in Slack. I then just had to explain the process I wanted it to go through and help it figure out how to call the correct API's to get the information it needs. It then created a cron job so it can monitor for new stories with new plans to implement. It fills me in on its progress in Slack. Simple tasks now turn into a small planning session to outline the intent, then I'm later greeted with a PR to review.

Github Copilot SDK now has native Plan Mode and it surfaces question choice types. I've improved my integration of Plan Mode into our project management systems Story edit. Plans can be attached to the Story and then pulled into a code agent using MCP later.

Cleaned up the user experience a little bit. Added support for viewing mermaid diagrams and improved the language for more non-technical users.

Trying out the new Github Copilot SDK which uses the CLI under the hood. This is cool because you don't need to do your own agent orchestration and tool call setup. Whatever the CLI can do, you can do. Similar to a Claude Code with a custom frontend. This is a very rough proof of concept idea of exposing agent plan mode in a project management frontend.

@Strategy STRC is a modern savings alternative designed for people who want their cash to earn more than a traditional bank account, backed by a digital asset you can’t print.

We want to see how you explain $STRC. Share a creative 15-second video in the comments below. 👇 Top submission wins a free pass to Bitcoin for Corporations at Wynn Las Vegas, Feb 24–25!!

Wrote my first Agent Skill. Well Cursor did. A documentation and change log Skill. I'm trying to think of quality of life or monotonous tasks that agents could do. This was the first idea, but I'm also thinking UI improvements next. Improving spacing and layout after major features are built. I started by asking Cursor that I wanted to create a process that reviews changes and makes continuous updates to a master documentation file and that I also wanted to create a change log that uses the git diffs to summarize changes. This led it to create a Cursor Rule that would remind the agent when to trigger the Skill. Then it created a Skill that is designed to know what file to update and how. I later asked it for the ability to initialize the documentation file if one wasn't already created. The video is the results of the initialization on a project that is in progress.

Software Development in 2026 Skills that are getting demoted: 1. Writing code 2. Shallow technical knowledge 3. Software abstractions Skills that are promoted: 1. Learning about Agents and AI. Continue to find ways to automate. Don’t assume we are locked into a single tool 2. Understanding the business problem and being able to clearly describe the solutions 3. Design. Product / UI / Architecture / Taste 4. Good Ideas 5. Testing and validating, and more Testing. USE THE SOFTWARE!

Our UI test project is set up to create and seed a local database for each test class. Each test class also spins up its own localhost site with a unique port. This allows for parallel processing without data conflicts. AI is great and learning your patterns so once you have some tests working, asking for more is a minor ask.

Automated testing has always been a love/hate relationship for me. It makes sense to do but it took so much time and thought that it often got ignored. Now that’s also changed with AI. I can finish a feature and then just ask for tests. We opt for UI integration tests using playwright. Once you have the framework down, adding new tests is trivial.

@aye_aye_kaplan Will support for Azure DevOps ever come?

real friends don't let friends go without Bugbot

@mntruell Remote execution of agent chats to local cursor instance. Not hosted agents.

Cursor seeks to be the best and most powerful way to code with AI. What are the ways in which we could be better?

@mntruell I want a solution for external users / non devs, to collaborate on a plan. Preferably without low level code language. And lots of interview questions.

When a support request is sent, it creates a ticket in our custom management system with those details and files attached, where we have an MCP server set up that allows the developers to reference the ticket or story and have the Agent automatically pull in this context. The Agent sees the files, downloads them, reads in the text if they are PDFs, and now you have a head start creating the solution. In my prompt I'm using a Cursor command to hide a bigger prompt that describes that I want to use a specific MCP tool and server I have registered with Cursor and if there are files returned, download them and read them in. Pretty wild.

I've been thinking how we can continue to close the gap between a user reporting an issue or creating a request, gathering the correct information, and then getting that into a prompt to generate the code. It now starts with a custom interface in the app, where a bug/feature/question all have its own unique questions. This in turn builds context including browser and user information along with screenshots and even optional user selection of the elements image and html.

I was inspired to create a similar experience in my workplaces custom project management system. Still amazing what you can do with a few Agent prompts.

Cursor.com/2025 Pretty cool year-in-review usage stats. I have 2 because I switched accounts.

Compound Engineering is what happens when agents write 100% of the code. At @every, engineers don’t type code anymore. They orchestrate agents. The shift: - Coding is no longer the bottleneck - Planning, review, and learning loops matter more than syntax - Each feature makes the next one easier to build The 4-step Compound Engineering loop: 1.Plan – Agents research the codebase + best practices and produce detailed plans 2.Work – Agents write code, tests, and iterate using real app simulations 3.Assess – Humans + AI review from multiple angles (security, performance, overbuild) 4.Compound – Lessons learned are stored so future agents never repeat mistakes Complexity still grows, but so does the AI’s understanding of the system. Result: - One developer can now do the work of ~5 - Products run by single engineers serve thousands of users - New hires instantly inherit years of institutional knowledge Engineering is no longer about writing code. It’s about designing learning systems that compound. — Thanks @danshipper and @kieranklaassen for sharing your approach. So much to learn and takeaway from this compound engineering approach.