So in 2026 as I write this, AI is intrusive to the point that everyone in Management want it because it is believe to work 10x and reduce costs or because everyone is implementing so must we.

And the predicament doesn’t stop there – it continues expand to a point where plans are chalked out to run a department entirely. And organizers scramble to solve the problem directly by introducing AI as a form of “automation” rather than “augmentation”. I pondered on this a lot and this is what I think of it as it stands today.

Although most organizations are way ahead with working AI solutions already and are monitoring for production performance and tuning. But this blog is focussed on someone who’s just arrived to AI conversations and have burnt their fingers in some POCs maybe. Here are my two cents –

Foundation of AI for Business Applications

Here’s how you start – and it’s not building your first agent directly. But are the systems hygienic for your first AI agent to not struggle? Here’s what I mean by that –

1. Clean your data – Make sure missing Sales or Operations data isn’t cluttering your system, giving unclean and unreliable reporting. Does management completely trust the reports you run today? Get the data right.
2. Connect your systems – Integrate systems together for not everything – but enough to make sure you don’t duplicate, miss data or rely on manual labor prone to making mistakes. If integrated, make sure integrations work well.

Now extrapolate this at the Org level and you already are starting on a high-trust metadata and data at your org for any AI agents to make sense on.

How to introduce it?

Now that data is clean and systems are connected –

1. Start small – introduce AI in micro-tasks of your workflows. Use it to generate what your team does obviously. The routine, mundane updates. Repeating tasks which can draw patterns.
2. With those patterns – form the knowledge base of the AI. Improve AI to feed off of the knowledge base itself. It could be a structured prompt engine or a library that the model has to traverse, process the documents and make sense.
3. Don’t use it for decision making – use it yourself so you can make a decision – just quicker.
   

Why do we feel it is not what it is meant to be?

I started off by thinking in a straight line –

1. I want to get to point B from point A – so, let me implement it and give it to the tech team to do it with some instructions.
2. Well, it does work – most times. But I’m still not happy with it. And here is the reason –
   1. No matter how perfect the answer is.
   2. Your judgement in that moment will override always — a confirmation bias or a change in situation suitable at that point in time will make more sense.
   3. You tend to take your decision overriding what AI has to offer for you (or rather recommend).
3. And you dismiss it saying – “It doesn’t work”.

Well, it does work – but it’s giving you an opinion so you could extend yours. It’s not meant to give you your answer, yet. And this is true for most cases if not all.

It’ll be interesting to see how this blog either ages like milk or like wine in coming years. 😊

With this, I kick-start my AI Thought Leadership as a new section and running commentary as I work to get our customers and partners ready and thriving on AI in coming years.

Hope this was helpful!

Thank you