Ai over the last decade ….

All of you have used or have been using AI for close to a decade now. Sometimes you understand that it is AI and most of the time you do not. In the last 2 or 3 years, AI applications have been so ubiquitous that we now imagine things happening for us with Ai instead of directly doing them. 

So, here are my observations about Ai over the last decade. When I mention decade, it is 2016 to 2025 as I write this in 2026. 

• A recommendation system based on a list is where Ai started in the last decade. Over the decades we have structured data which exists in the form of a table and recommendations can be generated from it. This became a popular AI use case.  

• Summarisations is another use case which took place. In 2025, think about how the advanced meeting notes agents work. They can transcribe the meeting, figure out what are the action items per attendee in that meeting, send each one an email with the action items and so on. This is very productive because you need not attend each and every meeting and still be aware of what is happening in the meeting through these tools … 🙋 

• Parsing of documents is another use case which is picked up. The document can be a contract, legal document or even a candidate resume. 

• I have worked on forecasting (prediction) of website traffic with Time Series models based on historic data including the seasonality. These use cases help the user build trust with the system and give confidence. 

• Another popular use case which I have seen working very well is creating emails/ messages for Users in multiple languages based on the User input/ natural language. This is GenAi with chat gpt enterprise API. Because it is not possible for a user to have read/ write knowledge for more than 2 or 3 languages. Some may know multiple languages, however it is not a frequent occurrence. AI in this case helps the user read and write content from multiple languages and help the customer at hand … 💰 

• Last but not least, I would say the use case of the decade is search where the intent of the user is identified with multiple sources and a relevant search result is provided to the user. This has kind of revolutionised how e-commerce sites drive traffic to their product pages and increase visitor to buy conversion.     

I believe more promising use cases will emerge as we progress forward. 

Here is a relevant and refreshing quote for you - 

“Life is like riding a bicycle. To keep your balance, you must keep moving.” 

―Albert Einstein

The process of development of AI - 

I believe this process has remained mostly the same in the last decade. It included idea, customer validation, ideating the feature, guiding designers to give on screen updates, giving the technical details, releasing, assisting the customer/ b2b users in adoption and repeat. 

In 2024 and 25, there are many tools which are aimed at improving how this process can be faster and made error free … 🏆  

Approach for AI onboarding - 

Enterprises also need an onboarding approach which helps them scale the product to the entire company.  Once the AI, Saas features are released in production/ to the customers, we need to handhold the customer users, give them webinars to explain the feature set and answer their questions … 👏 

The training data is of key importance. We need to understand the training data with statistical analysis, variances, etc. If the data is not normalised for outliers etc. it is not possible to get the best outcomes on screen for Users. We need to work hand in hand with the technical, Data Science and ML teams throughout the process. Making the AI work for an enterprise is a journey in itself. I will write about this in detail soon. 

Build vs third-party API - 

It is tricky to determine when to use AI features, APIs, or custom architecture. This is similar to that of Build or rent. If the need is urgent and users will get happy about these features and we will get more revenue; I have always suggested to my clients that we rent someone's API and pay for it. Eventually in the next 6 months, we can build our own. 

However, keep in mind that from a security, privacy and intellectual property perspective it is always advisable to own Ai in the long run. Cost, hosting should also be looked at. 

Think about becoming model-agnostic this during scale. The objective behind becoming model, platform agnostic is control. We should be able to manage our system outcomes with third parties. This approach is a bit expensive and time consuming, however it guarantees business stability … 🍎 

Similar analogy as that of multi, hybrid-cloud solutions.  

I believe 2026 onwards, new age AI, Saas tools will become more stable, relevant, and  deeper from domain perspective. Happy reading. 

PS - a picture I like very much. Good memories .

……. 

Swagat Irsale is Growth Advocate. He works with startups and scale ups to grow revenue and build products which enterprises love to use.