Technological progress is circular in what it rewards, even as it is exponential in what it enables.
We’ve spent centuries building technology to free ourselves from the limitations of human cognition, coordination, and physical effort. In doing so, we built economies around specialisation, institutions around coordination, and increasingly sophisticated systems to make the world more predictable.
Yet I suspect the closer we come to automating these capabilities, the more we will find ourselves returning to the qualities that made humans capable of flourishing long before modern technology existed.
Judgement. Adaptability. Relationships. Community. Agency. The ability to navigate uncertainty and make good decisions.
There is something fascinating about how post-AGI society will come to resemble aspects of pre-industrial society.
And therein lies a paradox.
The very capabilities that become most valuable in a post-AGI world are the ones we risk losing on the way there.
As we delegate more of our thinking and decision-making to machines, we surrender opportunities through which we develop these capacities in the first place.
And so, two seemingly contradictory futures emerge.
One in which intelligence abundance produces an extraordinary renaissance of human agency. And another in which we become more capable than ever through our machines, while becoming progressively less capable without them.
Perhaps the greatest challenge of building increasingly capable machines is ensuring that the humans using them continue to become more capable, too.
Acquisition and retention: after setup, what brings someone back?
In 2024, whilst building Ario, a personal assistant for working parents, we grappled with retention. We could get a parent to connect their calendar and Google account. But still never become part of their routine or take hold as a habit.
Anyone who has built a product, especially in consumer, knows how hard it is to get someone to try something new. Let alone trust it with their data.
Parents connecting their accounts signalled strong interest and trust. But would Ario come to mind when a school email arrived, an appointment moved, or they needed ideas for a weekend with their child?
I suspect many assistants (and their builders) are grappling with a similar retention question today. I return to it often as I try products like Grok, Muse, Instinct and Dots: “When, in my actual life, would I reach for this?”
It is on the user to remember the assistant, recognise when it could help, and decide whether using it is worth the effort. For a product promising less mental load, that is quite a bit to leave to them.
Between what we learned at Ario and what I notice using assistants today, a few things have changed how I think about retention:
(i) A general-purpose assistant still needs a specific first job. Start by giving users one recurring task to rely on the assistant for. For a parent, that could be keeping the calendar up to date when school plans change. Each time the situation recurs, the assistant gets another chance to prove useful. Earn consistent delegation of that task before treating a broader set of capabilities as progress on retention.
(ii) How work reaches the assistant is part of the product. The school email is already in the inbox. If using the assistant means remembering and opening another app (or iMessage), we have added a step at the moment we promised to remove one. Our Email-to-Ario feature had low adoption, but those who tried it used Ario more consistently and were among the users we retained best. They may have been more motivated to begin with; we could not conclude that email caused better retention. But it gave us a reason to treat the handoff as something worth building around. Make that first job easier to hand over before adding another use case.
(iii) Delegation has to beat doing it yourself, including the checking. Moving an appointment manually has a known cost. Asking an assistant means explaining the change, checking the result, and perhaps fixing it. If the parent still has to supervise every step, we have left much of the mental load with them. Judge the assistant by the total effort it saves, including what happens after it says the task is done.
(iv) A user who does not return is a question to investigate before it is a problem to fix. Perhaps the need never recurred. Perhaps it did, but the assistant did not come to mind. Perhaps they remembered it and chose to do the task themselves. Or perhaps they tried it, disliked the result, and saw no reason to try again. Start by finding out what happened at the next relevant need. Until we know that, another onboarding change or reminder is a guess at which problem we are solving.
(v) Retention should follow the work. Someone might stop opening the app while the assistant continues to help through email, messages, or a recurring task. Repeated reliance on completed work is stronger evidence of assistant retention than repeated app opens. Did the next relevant task reach the assistant? Was it completed with little effort from the user? Did they keep trusting it with that job? Build for those behaviours and measure whether they persist.
Getting someone through setup earns us the opportunity. Retention comes from doing the job well enough that, the next time it needs doing, they trust us to take it off their hands.
BREAKING: JOE BIDEN WINS
Joe Biden will be the 46th president of the United States, CNN projects, after a victory in Pennsylvania puts the Scranton-born Democrat over 270 https://t.co/pZ8pr4cMTG #CNNElection
It’s a special day for me & I want to share some imp news
To get children of sex workers out of that hell, I am starting program “PANKH” with 25 children & I’ll look after all their needs incl shelter & edu! I urge others to come fwd & contribute too!
EVERY LIFE MATTERS!
Why are we seeing devastation of this cadre? Such negative news, horrible situation in life and two greats passing on... @irrfank and now @chintskap#Unbelievable and #heartwrenching@chintskap you left such a lasting impression on me with that small talk!
@anubhabhonsle Hey @anubhabhonsle would love to chat to you as we are a small group come together as a #thinktank to help people via upskilling/multiskilling. Would you be keen to connect??