ctrl+r #08: 2026's AI wake-up call, voice-first workflows, and the data engineer identity crisis
From Karpathy's grief cycle to Hyprnote workflows, plus why companies are locking down their APIs
đ§ 2026 is when you canât ignore AI in your work
Iâve been playing with AI since 2024. I remember trying out Cursor for the first time and being mind-blown by how good it was... but it was actually BAD in absolute terms. It could only edit one file, and code quality was meh.
In 2026, we reached a tipping point where AI is so goodâespecially at codingâthat you canât ignore it in your work anymore.
Iâve always experimented with AI and different workflows, mostly around prompts and some limited MCP services. In 2026, I need to master more of these workflows.
Specifically, Iâm interested in diving deeper into:
Building various MCPs for my daily work
Working with multiple sub-agents and pushing work async as a reviewer
How to manage memory for AI
Some powerful alien tool was handed around except it comes with no manual and everyone has to figure out how to hold it and operate it, while the resulting magnitude 9 earthquake is rocking the profession. Roll up your sleeves to not fall behind. - Andrej Karpathy
đ ď¸ More voice, less typing
Iâve been experimenting with voice recording tools, specifically for:
Meetings where Iâm not in control of recording notes
Meetings where I donât want to send the AI summary
For myselfâspending more time speaking than writing
Not having a big red button âRecording in progressâ
I think the exercise of writing is still important, of course, but thereâs something nice about speaking freely and getting a summary or reordering of your thoughts.
I stumbled on two nice open-source products: meetly and hyprnote.
I went with Hyprnote because itâs just markdown (yeah!) and I love their philosophy on not wanting to compete with tools like Obsidian, as described in their latest blog post âFilesystem is the cortexâ.
Iâm still figuring out my workflows, but all in all, itâs working pretty nicely!
đ What I read/watched
Welcome to Gas Town: A rather crazy take on how to run multiple agents with tmux. Note that Iâve seen more people managing agents through tmux lately.
Full Course: The AI Stack We Actually Use for Prototyping, Strategy, and Personal OS (2026): Another example that thereâs no manual for the best workflowsâyou need to figure it out. Good examples here using text files and markdown.
Defense comes to software: Tomasz mentions that a lot of companies are restricting their APIs to own the full stack due to the rise of AI.
Data Engineers are going to be whatever they want to be : Good take from Sungâs Substack on how the role of data engineer is shifting into multiple subroles given the rise of AI.
Last saturday was my last day of skiing with full powder before going back to work after my parental leave!
