Imaging that you are a hiring manager. You open your laptop on a crisp Monday morning, energised, and keen to firestart the week. The startup you work for has allocated some budget, and needs you to define who gets hired and at what compensation. You want to dig into the data; visualisations always get your creative juices flowing. You realise this is when you most need an AI assistant. 'Hiri, show me the org chart,' you want to be able to say, and obtain a beautiful map of the teams and members that you can zoom in / zoom out and hover to see all relevant details.

To identify the teams that really need a resource you say, 'Hiri, pull up data from the project management system and highlight the teams that are struggling with their deadlines lately'. Ah I see. Are these the same time which have raised hiring requests? Can you show me? Hmm 'Can you show me the aggregate skillset for three teams I have picked? 'Hiri, help me understand how the skillsets of these teams are different from the skillsets of our best performing teams' ... and so-on, you continue through the morning, sipping on your fresh celery and beetroot juice, unimpeded, hoping to make a proposal of the 'who and at what salary' by the same afternoon.

Now let's zip over to the backend to understand how such use cases could be powered.

Traditional SaaS: An assortment of features on a web dashboard. Plug in your HR data and start shooting. Click 'here' to 'load data'. Click 'there' to convert it into a 'chart'. Click 'another-place' to load the skill sets. Click 'some-other-place' for summary statistics on teams, and so forth. While traditional SaaS has been super handy, it has two bottlenecks - (a) features primitives provided on the software are limited and static, (b) to get to the final albeit approximate solution, one has to climb a learning curve to know the ins and out of the dashboard. (hint: the better you want the approximation to be, the harder this curve).

Over time, the dashboard is cramped and sluggish and you are stuck in a loop using the same handful of features you've always used.
Agentic AI - the FM way: Would you like working with an assistant who's by-and-large sincere and honorable, but just sometimes makes up facts, adds and removes employes from the roster, gives them skills they don't have, and doesn't tell you when it is unsure of itself ? FMs are good, no doubt, but more as sparring partners than as an abiding assistant. They need to be challenged, critiqued, and asked for specific explanations/evidence which the user, aka you, needs to verify (without becoming complacent).
What's a good balance? Agentic OS: A self-constructing kernel! You coudl say, 'Hiri, make a chart with the organisation data so I can see the hierarchy of teams.' and the OS will determine and fire up the appropriate code-unit (pre-written and pre-tested) with a customised set of input parameters. You can operate it with natural-word instructions (verbal or handwritten). And it provides a CoT reasoning as it solves for the task, 'I am using this parameter from this data, and this other one from here, and now I am sorting by this key, and plotting with this color scheme.'
But what if the code doesn't exist? Well, just like a computer tells you when its missing a certain software, or a plug-in, this agentic OS tells you when it doesn't have a feature in its arsenal. But it doesn't stop there. It creates the outline, raises a ticket and sends it to dev.

What do you think?