Pondering what it means to be a human working with AI Augmentation
Human edited and authored, with a small amount (~10%) of AI input (mostly product framing)

I just got done with a 7 day sprint at Decibel, working to harden and market-ready our upcoming suite of software, Tools.
The workflow is so powerful, the results so rewarding, I found myself pouring 14-16 hour days into the effort. Almost compulsively. It was one of those things where I knew something was different // felt different, the results were different but I couldn't quite verbalize what was happening. Intoxicating, to go from idea to working feature in hours or days.
Well I've landed on a metaphor - we'll see if it resonates. As I said in my last article, AI is more than just a copilot, and that still holds true. That metaphor is the Infinite Stack.
The Infinite Stack
I think the key is this: plugging into a fully AI augmented workflow is freeing in ways we as humanity have never had before. Infinite stacks to pull from and rely on. If you don't know something, you can pull it. I'm immediately, almost begrudgingly, beset by the comparison to Neo in the Matrix - loading up his next set of skills. Seems like a joke, but it's not.
The 'trick' is learning how to navigate these systems so they are compounding on each other and with you. Worry about the intent, the goal, the architecture, the structure, the problem, the 'feel' of it and pull in stacks as you need them. There is an infinite reservoir of knowledge you now have access to. Use it. No need to get overwhelmed, you can offload things like project memory.
If done correctly this doesn't turn you into a drooling moron - it turns you into an augmented human, now with infinite reserves of compute.
People have fretted, including myself, that by offloading work to AI will somehow stall our growth and blunt our intelligence. The opposite I believe is actually true. You are suddenly 100x more capable now if you figure out what works for you and harness these tireless reasoning machines to do your bidding.
The Challenge of Context
There are challenges today, sure, like Context, with a big C. Instance based intelligence has its pitfalls. What were we doing again? Why do my stacks forget basic things like what folder we are working on and what file we just had open. These problems are transitory and Decibel Tools helps fill the gap in the meantime.
Realizing that you have this Infinite Stack to draw from is freeing - identify what you want to offload to AI, come up with a plan, workshop it with your model of choice (in my case ChatGPT is working the best for big picture planning, the added memory capabilities really come in handy when planning across projects, and I still often do a planning step with Claude when the plan involves the codebase in some way, which it usually does). Rich has a different mix, and it works because we each found what resonates with our individual working styles.
Making Knowledge Legible
This is another phase shift - for the Infinite Stack to work, all project knowledge MUST be legible to your AI team, not just your human team. If not, it doesn't exist. This causes some amount of overhead, and if done incorrectly it could feel insurmountable with headaches and accumulating problems - just as, when done well, friction seems to melt away, if your system is disconnected or has major gaps the problem can compound and your AIs are suddenly Lucy in the cake factory fumbling to keep up.
So, the Infinite Stack is not a given. It is a state of being achieved when you align your systems to communicate effectively, where humans and AI work in concert, and project knowledge is properly maintained, tracked and tended. We've started thinking in terms of 'health' at Decibel. Is our system healthy? Is our knowledge accurate? Doing this allows the LLMs to run and not just be pure, 'write me this one method please and don't mess anything else up'.
The Human in the Loop
Now that you have a team of tireless AIs to help, let that work for you and not against you. The human in the loop is still an absolute requirement both for inspiration and evaluation. And given the state of the models today, a certain amount of cat wrangling. This will recede over time.
I was so dialed in this week, partially supercharged by the Claude Code research preview, that I noticed every little update they made to the experience as they did it - 'oh ok they are compacting threads now', 'ah they have added cloud connection status'. I assume AI companies are already working this way and are in a flow state of software development. With Decibel Tools, so can you take full advantage of the Infinite Stack.
The Four Avatars: Decibel Tools
A bit about our stack, I've dubbed them the Four Avatars of design / dev:
Decibel Designer
Captures key creative details so the team can focus on the work. Design tokens, visual language standards, brand standards, design systems, even things like principles of design and moodboards go here.
Decibel Architect
Our version of systems design memory. Architecture decisions, boundaries, contracts, project or even company level methodology is tracked. Architect is where we teach the stack how to build.
Decibel Sentinel
Provides forensics and visibility. Sentinel watches repos, scans commits and provides an LLM augmented report of project health, hotspots, tooling, tracked issues and gives recommendations.
Decibel Oracle
The newest member of the four, the goal with Oracle is the queryable mind of the system. Watches across-company and across projects. LLM project review and guidance on hotspot issues and potential roadblocks in development.