I have a hypothesis. Hear me out.
Where we will see the biggest benefits of AI will be small teams measured (and paid) based on delivered outcome.
Knowledge work, is defined by it’s non-routine and creative nature. Not two tasks are the same and the way get to the outcome have to be figured out. As oppose to manufacturing line, where the tasks are well defined and isolated.
Due to its nature it is hard to measure. It’s hard to measure if Task A and Task B are the same even if they sound similar and if Task B will be completed in roughly the same time as Task A.
In technology space, companies have spend decades trying to figure out how to measure throughput, forecast delivery and manage performance trough numbers. There are ways how to get indicators, however it’s impossible to reliably measure performance of individuals or teams. To make measurement and scaling a bit easier, we have seen a lot of companies to move more to manufacturing model. A knowledge task was split into multiple tasks – analysis, development, testing, management etc. These then got their specific job roles. Business analyst, software developer, tester, manager.
Then we can have a look how things are progressing trough this pipeline, how long does it sit in a queue and where the bottlenecks are. There are still quite a lot of caveats in this as it requires a LOT of discipline and expertise to be able to manage this ways of working well.
A key part of this ways of working are “formal” handovers. Such as analysis is done and ticket can be officially picked up for development. Analysts work is done and developer should have all information needed on a ticket to get their work done.
Problem is that it’s never that easy and that’s why Agile methodology became a thing. Get the team closer together and remove some of the formalities. The challenge for a lot of organisations is that it’s then hard to manage as you are looking at output of a team rather than individual tasks closed/moved.
With remote work we’ve seen insurgence of ways how to try to quantify and measure knowledge work to decide if remote workforce is performing as well as the one in the office. Quite often, being in the office at the desk became proxy measure for staff output. With remote work that proxy measure became green light on teams and responsiveness to chat messages. It’s the vibe based approach in absence of the quality of manufacturing style management.
Where does AI come in this is that LLMs make people much better at being T-shaped (multiskilled). And the T thickens with AI.
This will lead effectively to not just moving teams and people together, but in some cases merging the roles as well. When the iteration between analyst and engineer isn’t raising tickets, moving it in queues, chasing it, misunderstanding, reiterating, waiting (at worse case); but a chain of thought in ones head with iterating against LLM.
And when companies are looking at the AI benefits in the current structures, roles and processes it will always deliver much less than they expect and changing that model will difficult. Instead of AI genuinely helping boost efficiency and improve quality of work it will be used to prepare task for a handover. Still useful, but much less if the handover wouldn’t exist.
The challenge from management and performance monitoring is that the task won’t have formal handovers. People won’t be changing status “In analysis” and “ready for development” 10 times a day, because whilst they were developing they had to go and do a bit of data profiling again.
That’s why I believe we will see the biggest benefits of AI will be small teams measured (and paid) based on delivered outcome. I think fixed-fee pricing in different services will become more common practice as it will be a bet on that 3 experienced T-shaped people will be able to deliver project cheaper than how currently projects tend to get delivered.
If you are looking at a small to medium size data project and your list of “resources” looks like this:
2x Business Analyst (we need someone who will understand what is it that we need to do)
0.5x Solution architect (to design infrastructure)
0.3x Data architect (to design the data related services)
0.5x Data modeller (to design the data model)
1x DevOps engineer (to implement infrastructure)
2x Data engineer (to develop pipelines)
1x AI engineer (to hook up LLM)
0.5x Software engineer (APIs are not done by anyone else)
1x Software tester (only person paid to verify that it works)
0.3x Web developer (for a simple UI, but we don’t know when we need them)
0.5x Data analyst (as BAs and DEs don’t do dashboards or data profiling)
1x Delivery manager (A lot of different people and handovers need a lot of management)
0.5x Scrum master (too much admin for DM, needs support)
0.25x Account manager (with so many people on the project, so many handovers and growing costs you have to have someone who is paid to pay attention, right?)
You might have a bad time.
Just above average 