As always – these are strictly my personal opinions
Let me start by reiterating that I still don’t agree with Satya Nadella about Apps just being nothing but database CRUD
So where do I see apps going next as AI gets better and better ?
Let’s take an HR platforms as a starting point. They cover a wide array of workflows that deal with finding talent to onboarding and training them to managing performance to paying people and ensuring compliance and separating people from the enterprise. Probably many many more such processes are in scope of such systems.
Now fast forward a few years when Agentic AI moves from science project to active deployments. Now you literally have the same job done by both AI agents and human workers . An example would be say handling invoicing where AI agents do the simple ones and humans do the corner cases. It won’t be one agent – it will be hundreds or thousands of agents who are doing all or part of the enterprise workflows.
That means you now need a system of record to recruit the best bot , onboard those bots, train and retrain them and so on. A science project doesn’t need any of it – but a scaled deployment absolutely will delve into massive chaos unless it is governed.
HR systems of the past have massive technical debt already. If they try to tweak their metadata and data and UX to accommodate digital labor, it will take quite some time and money. This opens up a massive chance for startups with no tech debt to create simpler platforms built for such co-existence. Knowing my exceptionally talented friends at the big enterprise software companies – I am sure they will find clever solutions to all of it, but the threat of disruption is quite real for established app companies in my opinion.
HR was just an example – literally every workflow in an enterprise like marketing and supply chain and so on will get redesigned to make use of digital labor (and other conventional AI goodness too – but probably those are incremental and not as disruptive).
AI will help write and test a lot of software – but not all apps can make use of them equally. I am not sure if apps that need a lot of rigour and consistency like accounting will let go of human control and let AI take over coding and testing.
That’s just the software development side of the story.
Think about the FinOps side of the story next. Today – where most of enterprise workflows are deterministic – it’s quite hard for most companies to plan for and optimize their spend on apps sitting on cloud. There is a whole genre of very funny jokes about hyperscaler bills on the internet. Now think about what happens when AI with its probabilistic and compute hungry nature becomes mainstream. That whole discipline will need to be redesigned !
Let’s also think about the SI ecosystem that is vital to deploy software. When SAP and Oracle and so on got challenged by new app companies – something that was touted loudly was that these newer platforms won’t need significant SI work. On quite a literal sense – that might be true. But if you look around – all those companies ended up building very high SI ecosystems around them.
So where does SI ecosystem move to when apps get disrupted?
I think BPO will be the first casualty – where most of the repetitive work can be automated away. There are SI companies out there who have hundreds of thousands of BPO employees in a labor based model. They all have smart engineers too – but since public markets don’t take kindly to drops in revenue, they have limitations in massively automating. That leads to two possibilities – newer and nimbler “tech first” SIs will go after the incumbents and win OR a hyperscaler or software vendor will wrap the labor into an outcomes based contract with clients and just dis intermediate BPO types services. Either way – BPO the way we know it today is going to be toast. Other things like production support (AMS) also will go this route.
But the more interesting question is whether software deployment can minimize its dependence on SI firms at all. So let’s delve into that a little
AI is already good enough – or close enough to be there soon – to not need a lot of human labor for creating reports and forms and so on. GenAI is quite effective with data manipulation and soon we won’t need as much human labor for data conversions either.
That means the big question is whether developing interfaces and enhancements can get easier with AI. Enterprise software is a lot better today than 25 years ago – but interoperability is still not a key strength. Metadata, APIs etc are all different across platforms – and given most big platforms grew by acquisitions, often they are not very well standardised within a platform either. That’s what has historically needed a lot of skilled SI labor to implement software. GenAI does give options to change this equation drastically.
Even if development work itself still needs a lot of skilled labor – think about things like discovering logic of old code, testing and so on which today take a lot of time and effort. If you look at a transformation project end to end – those are the things that eat up time and money the both. They can absolutely make use of AI to help make massive productivity gains
Just like with software apps – it remains to be seen if SI companies will disrupt themselves or get disrupted by new entrants.
There is some net goodness in all of these things if we are adaptable. Those AI models still need to be trained to take over repetitive tasks – the BPO folks doing repetitive work today might be quite valuable in training AI to do such tasks, and then move on to higher order work like orchestrating workflows as market changes. It’s not a zero sum game unless we make it one by sitting around waiting.
For all of us in tech world – the choice is between getting excited and learning and adapting fast OR getting paralyzed with fear, not learning and just reduce our economic value as market rapidly changes around us.
And Vijay Says... 