TL;DR:
In this post, I remind myself about 6 traits that matter in the age of AI:
- Intellectual Agility — rewire how you work, constantly
- Curate like Crazy — less is more; avoid AI fluff
- Second Brain — build a compounding system of context
- New Muscle Memories — pause before doing; ask "can my agent do this?"
- Being a Good Manager — of agents, not just people
- Being Human — empathy, agency, and curiosity as the last moat
Recently this AI thingie has graduated from being a fascination to being something close to an obsession. I have a unique vantage point where my products are powered by/powering AI while I also grapple with how AI can actually impact 'how i work'. In this context, like so many of my professionals I am super fascinated by the endless stream of innovation. Every weekend is filled with constant tinkering, wonderful POCs and nice demos. But there is one strand of thought which has been recently keeping me up. Fully Realising the magic and power of AI at my day job in office - To a point where I can offload a significant part of my work to it. To a point where i can automate large parts of my current work and make my current workload kind of new baseline table stakes. I don't think my aim is to do less hours of work, but it is to honestly explore the actual limits of the technology. If this is true, hopefully it will allow me to take on higher-order activities, expanding my toolset, achieving a higher level of performance at my work — generally expanding the area of impact — and fully realising my individual expression.
It is also quite the cynical part of me as i try to answer these limited questions:
- Does AI just make us 'feel productive' or does it actually increase our individual throughput
- Yes we can see that AI is slowly doing longer horizon tasks - but can it yet replace a human worker - maybe a junior employee to start with?
- At what point in time, will the major model providers, unshackle the models and make them available as "autonomous agent employees" - with their own structured onboarding program, RL loops, reporting/feedback regime.. How will the workplace look like?
I have consciously wanted to focus on a limited set of things in this blog post - I have decided to focus more on what as an individual i can do. I have left the wider, scarier societal implications to other people. This article is just a note to self - on how i feel one can navigate these uncertain/scary/perilous times.
I have tried to boil this down a set of traits that are important. Some of them are nothing new but suddenly becomes more important in the age of AI.
Intellectual Agility to Shift and Change Workflows
The workflows you're proud of today are the ones you need to be willing to kill tomorrow.
We are all attached to a particular way we show up to work. The way we use Email/Slack/Deep work/handle notifications/handle meetings etc. But i have noticed recently that how i am working is actually changing. Vicerally/Physically. And everytime I am forced to change this because of some new innovation/tooling, I notice a strand of resistance within me. Not being attached to how I have been doing things earlier is important. In fact, constantly seeing how to reinvent or change or prune some of those workflows as new tools, capabilities, and approaches keep rolling out. It involves being honest about what I can truly and uniquely contribute, and being willing to table the rest to the Agent.
Human Curation is the new Organic
In a world drowning in AI output, the rarest signal is a human who chose to say less.
Having a human touch will be noticed, valued and maybe even encouraged. Optimize for connection rather than volume in our work output. We generally have the instinct to show to the world the sheer quantity of proof-of-work. This could be lines of code, very verbose reports/PRDs with too many unnecessary sections, etc. As everyone uses AI, this instinct to pile on — this is the most disrespectful thing to do to our colleagues who are bombarded by similarly fluffed up AI-generated content.
AI shop or AI baseline is now everywhere. And while the baseline quality of what a zero-shot prompt keeps going up, it will be so easy to confuse volume of content, work, and output tokens with quality of work or actual productivity. As someone faced with both baseline content (created by others) and content created by my own inputs, I need to emphasise my own critical judgment and discernment.
Building a Second Brain
Build a system where context compounds automatically, so every piece of work feeds the next.
I started reading this book called "Building a second brain" maybe two years back — yes, it helped me to some extent. Based on my reading, I moved all my OneNote glorious confusion to the PARA structure. Having a set of folders under 'Projects' gave me an instant sense of focus and prioritisation, and more importantly a bulwark against 'BAU' mode — making sure that I don't passively show up and just go through the motions. That was then.
Now, with Claude code by my side, externalizing useful bits of data from my brain for my agent has become a necessity. I constantly find myself manually creating context for every new project that I am taking up. The system of context is not growing organically as I work and move from one project to another. Recently, I have been hearing about Obsidian and somehow I bit the bullet and moved my notes from over 12 years from OneNote to Obsidian. This is still a work in progress — My goal is to setup up a system where memory naturally accrues and compounds over time. It feels like the last severing of my connection to Microsoft is secured — I have been a long-time Office/Windows user. So long, MSFT.
But so far, Obsidian + Claude seems to be a very nice unlock. One thing I love is that in Obsidian, the PARA structure in the Obsidian app mirrors that of the local file system on my Mac — which means I can add to the same system of context via both the file system and the Obsidian app. Right now the setup is very simple. I still need to resist the temptation to go down the rabbit hole created by Obsidian fanatics freely available on YouTube and Reddit, and let the natural requests and progression of automation guide the way. Having said that, I feel this system will slowly grow in complexity.
The main reminder here is to make sure every unit of work, every ad-hoc communication/meeting minutes/decision/insight adds to this system of content — and get that good old compounding going.
Identifying and Strengthening New Muscle Memories
The new discipline isn't doing things faster. It's pausing long enough to ask: should I be doing this at all?
One of the things I noticed that really helps me increase my throughput is to identify that moment when a new idea takes root. This could be an idea for a new product feature, a new experiment/POC, a new outreach to a partner team etc. And immediately the visceral reaction which follows me is to literally do a very quick cost-benefit analysis in my head, and discard that idea if it is too much hassle. Now, with AI, I need to consciously hold back and do another specific piece of evaluation: 'Will my agent be able to do this?' Do I already have content in my second brain which will instantly unlock this piece of work?
This is also true for work which i might have already done it a couple of times and is quite repeatable. But I find myself raring to go out and do it myself - to get that dopamine hit. Instead of taking a step back and thinking "Hey isn't this simple enough that i should rather spend time automating it". The muscle memory is to take a step back and honestly introspect whether I am adding genuine value here. Creating a repeatable process — which delays that dopamine hit but which lets me automate future instances of that work. I feel this is critical: the ability to iteratively take steps back and delegate mundane tasks to AI, so that one can fill the day with higher-order items.
I feel developing these muscle memories are important. Interestingly these traits are quite introspective. They require cold/hard look and asking tough questions: Are you feeling good because you are busy? Or are you genuinely adding value here?
To summarise:
The new evaluation checklist when an idea strikes:
- Can my agent do this already?
- Do I already have context in my second brain that unlocks this?
- Is this repeatable enough that I should spend some time to setup an automation/workflow?
Being a Good Manager
Treat your agent like a direct report — high bar, immediate feedback, push back when it cuts corners.
A weird, strange thing is happening. It is never a better time to work as a part of flat teams — working as an IC in a flat team, being hands-on, being close to the action and to customer voice and market feedback. At the same time, everyone has become a manager — a manager of agents. This is something I feel I am seriously lacking. The idea of being a good manager of agents is curiously similar to being a good manager in the good old days: setting clear expectations, having a high bar, giving good and constant/immediate feedback. Whenever i see some mistake/shortcut that Claude took, honestly I am not pushing back enough. I am still a kid in Candyland. I feel i need to push back and expect better things.
Being Human
We keep redefining humanity as whatever AI can't do yet — but some things may genuinely hold.
AI has forced us to introspect: what does our humanity mean? What does it stand for — what are its defining characteristics, its constraints? Surely not the Turing test (that GPT has passed a few years ago). Not intellectual capital either - these models already surpass us in bare-metal intellectual horsepower.
Recently i am thinking is it Emotional nuance? Is it the one which sets humans apart? Sometimes when i read a document which seems like it was created with AI, I feel a strange sense of betrayal. But honestly on the other hand, if I have to admit, relating with agents sometimes feels less like drama than some humans. So what is it? I honestly don't know. I feel it is quite similar to the "God of the Gaps" phenomena . In philosophy of religion, “God of the gaps” refers to the habit of invoking God specifically to explain whatever science has not yet explained, so that divine action is located in the remaining gaps of scientific knowledge. In a similar way, I feel there is a similar "Human of the gaps" going on - we keep shifting posts - The tendency to relocate claims of human uniqueness to whatever capabilities AI systems do not yet have, so that “being human” is always defined by the remaining gaps in AI.
But for now, I remind myself that a few things matter — things that one can act on now: showing real empathy and having that empathy with customers and colleagues. That human connection will be the new organic — that hard-to-define aspect of what it means to be human. I feel this is an evolutionary instinct and my bet is that people will continue to care about it and be able to discern it. Yes Agents can 'act with empathy' but is that real empathy? Or is the agent outputting tokens associated with 'empathic latent space'
The next thing which i feel will be uniquely human for some time is sheer agency - I know we are calling the current breed of AI as "Agentic AI" - denoting that these models are not pure-reaction - a la chatbot. But i feel there is a realm of agency which is uniquely human. Yes an agent can do increasingly longer horizon tasks that you can start defining more and more broadly. But i feel there is a unique alpha associated with the act of judging which actions are more impactful/meaningful.
The final thing that i feel will continue to be uniquely human is Curiosity and Creativity - I feel there is unique Alpha in exploring and following those random sparks in our head/body. Suddenly we feel it is important to pursue some particularly random thing - not because of our job. But I strongly feel those random sparks are not random - they are our minds and bodies leading the way towards what is needed for us at that particular time. It is hard to explain this bit without getting a little bit mystical. So i will stop with that.
I have tried to focus this blog post on what we can do which is in our realm of influence. I don't have answers to the big questions. But I know what I can control. Honestly, I feel this is a scary time to be a parent. I shudder to think about the bigger societal questions: employment/labour market, vertical takeoff, strategic dependence, Agents tainted with human malice/greed, Concentration of power etc... I have no idea who is thinking through this. I hope somebody is.
Maybe our humanity empowered by these agents will help us answer and explore these questions.
PS: I wrote this article on my uniquely scrawly handwriting on a very papery notebook. LOLZ.
Comments
Post a Comment