Adjusting org structure for AI
Goodbye gatekeepers, hello guardrails
Today: Departments as gatekeepers
In large companies, different departments often serve as the interface to various software systems. Software engineers own the codebase. Designers own the visual elements. A data engineer owns the database(s). HR owns the HRIS with employee data and policies.
Everyone is the gatekeeper of their own system.
If someone needs to access the information in that system, they must go through that department.
This evolved for logical reasons. Software engineers write code, customer support reps cannot. If a customer support rep finds a bug, they write a ticket for the engineer instead of debugging themselves.
With AI, all of these systems can now be accessed with English. No technical knowledge or understanding of a particular UI is required.
Even the best software engineers are finding leverage using English:
When any system can be accessed with a prompt, there’s no need to go through the system’s gatekeeper just because.
This means it’s time to re-think who does what at work.
The Future: Optimally allocating work
Cost
For any job to be done within an org, you want that job to be owned by the lowest cost resource and completed in the fastest time possible holding quality equal. If a lower cost worker can do a job more cheaply, you’re best served offloading that task to them so you can spend your valuable time on higher order work.
The market is catching on to this reality slowly but surely:
“It just makes a ton of sense from a cost perspective. Why would you be pushing work to your most expensive resource?”
— Head of AI at a Growth Stage Startup
Because English language interfaces open up access to skills that were constrained by expertise, those jobs can now be done by new teams or individuals within a company.
Speed
A second principle of allocating work is you generally want the work to pass through the fewest sets of hands as possible. This is why, all things equal, startups move fast and big companies move slow. Big companies are full of gatekeepers that you need to work with and “align with” to get simple tasks done.
Today, a single person can now achieve what would have previously required cross-functional input. This means the same work can get done faster.
We need guardrails for this to work
This is all great in theory but there are some practical realities that have to be considered.
It would obviously be chaos to allow everyone to have access to any system. Even for more benign tasks like querying a database, there’s almost always going to be data that not every team member should have access to.
Also, just because you can do something doesn’t mean you can do it well.
Technological guardrails
We clearly need some guardrails to make a new org model work and these guardrails need to, among other things, manage what information people can access and what they can generate.
Let’s start with information access.
Imagine, for example, you’re a PM at a startup and you want to understand what feedback Customer X has for the product. Rather than asking your customer success counterpart (a CSM) about Customer X, you type in a prompt to your Enterprise RAG solution. That RAG solution is integrated with G-Suite where the CSMs take notes. This is also where the CEO takes notes and, it turns out, there are notes from a conversation between your CEO and Customer X outlining a deal for Customer X to acquire your company. The content of these notes leaks into the AI’s response and now you’ve now been exposed to this highly confidential information. This is also material non-public information if Customer X is a public company. Very bad.
For departments to have direct access to information that used to sit with another org, there must be protections put in place. This fiasco would never happen if the PM sent a Slack message to their CSM colleague with the same question, as they would have pre-RAG app. However, 99 times out of 100 this new workflow is faster and less cumbersome for the team.
Next is the generation question.
Specifically, companies need to ensure that the content that is generated across their org fits within their brand guidelines. For example, if you’re McDonald’s and you want to empower more people to generate visual assets, it would be very bad if what they generated was this:
The red and yellow is iconic! Not so much when it’s shades of green or pink.
Org structure guardrails
While technology allows for the implementation of strict guardrails, orgs still need a sense of who owns what to function. These are softer but no less important.
Without any level of top down guidance, shifting work from the hands of one team to another is unlikely to be successful. And, seeing as we’re still so early in the AI cycle, there are no clear ‘best practices’ to rely on. The most forward thinking teams will need to come up with the rules as they play the game.
It’s going to take time to get it right
If you’ve gone through a re-org, you know that they can be painful. Lack of clarity creates friction until it’s addressed. In the worst cases, this gray area is never addressed and teams are constantly asking themselves “Am I supposed to be doing this or are they?” Then, when things don’t get done on time, fingers are pointed.
Worst of all, when re-orgs happen there’s always a group that feels like they’ve ‘lost’. Maybe because they have a reduced scope or have a perception that their new title is less prestigious. As roles change due to the new capabilities of AI, I suspect there will be a number of people who feel slighted.
Nonetheless, companies that are too afraid of this friction are going to be left behind. The potential productivity gains and cost savings from AI are too great. If you aren’t rethinking how you’re getting work done, your competitors are. Even if it takes a few iterations it is worth the up front pain.
I’m excited to see how innovative companies adapt their orgs to the world of AI. If you know of any, I’d love to hear about it!