Agents are changing the size and shape of teams and this in turn is changing the shape of the organisations they are part of. At SoftwareSeni we’re seeing this in our clients and our own internal teams, where smaller pods let everyone have more pods and pursue competitive advantage along multiple paths in parallel.
Agent usage shrinks teams. This does not mean adopting agents leads to a reduction in headcount. Why would you reduce headcount when you can multiply revenue-per-employee? Adopting agents leads to teams splitting into pods, and those pods, despite the single digit headcount, can match or surpass the output of the original team.
There are caveats. The biggest is the old programmer’s saw – Garbage in, garbage out. But in this case the garbage is the experience and skill of the pod’s team members. An effective pod needs effective people with the deep understanding of their field that allows them to accurately judge agent outputs. Because that is the bulk of what they will be doing.
With AI to handle execution, team members’ responsibilities spread across roles whlie shifting to a managerial slant that rewards expertise. Time is spent on planning and design, and then on verifying and correcting agent work.
While humans and their needs, particularly coordination and salaries, result in natural constraints on team size, agents can saturate any budget, allowing for a “headcount” that can grow and shrink depending only on how much you want to spend per million tokens per day and the difficulty of the work to be done. Some work requires the big models and the $50/M tokens.
The consequence of essentially unlimited agents is your pod’s managerial duties can saturate their ability to review, verify and decide upon agent outputs. Where a year ago decisions that shaped outcomes might occur at the beginning or end of a two week sprint, today those decisions can be stacked up in queues as agents log issues, work on issues, and make merge requests.
And those decisions haven’t gotten any easier. Your staff are getting less time to make them, and the pods, with their multiple hats per role, have eliminated opportunities to confer with team members working on the same feature or even with the same skill set or seniority. Everyone is now the sole expert on their slice of the product. Their role has been condensed, stripped down to the essential actions that create value: deciding what to do, setting production in motion, and recognising when it is done.
Meta attempted a full scale overhaul of their organisation with the goal of reducing many teams’ headcounts by 60%. This caused a backlash from Meta employees and the full transformation plan was abandoned.
But for some teams, like Facebook’s Feed and Reel team, the transformation completed. The result is fewer PMs and fewer layers. ICs are now direct reports to the VP Jaglit Chawla. In the past, there would have been two or more layers between the two. The result: as a flatter org, Meta is building more.
Chawla’s team has built and launched focused apps for sellers, creators, and group admins – ideas that had been sitting in the queue for a decade.
They also recognised the importance of expertise that is not app or even app building specific. Roles like data science and design work across pods as they are needed, blurring the line between strategic and tactical, between org-scale and app-scale decision making.
Roles might vanish, but the work remains
The use of agents, particularly agents provided with the tools and information to handle specific aspects of development, reduces headcounts by allowing the responsibilities of a role to be folded into the role that traditionally depended upon it.
This is most clear in developers. QA used to require a dedicated specialist. Now it can be handled by an agent overseen by a developer.
Teams no longer need dedicated Ops engineers. Again, agents can handle the responsibilities while overseen by a developer.
What needs to be clear is that the agents replacing the Ops engineer and QA aren’t created out of a thin air. Organisations need a Platform team that build the tools and provide the process knowledge that an Ops agent or a QA agent use.
Organisations have always built tools to build products. Now those tools include agents, as well as the tools agents use. And a small dedicated team, the AI Platform Team, or AI Enablement Team, needs to be in place to bulid them.
It’s not just team size, it’s also layer count
Management is the team between the teams, with the traditional responsibility of coordinating across teams and between layers. Drop headcount and you have less coordination overheard. Below a certain number of people management is a burden.
At SoftwareSeni, we’ve found Project Management roles are now rarely needed. Our pods are lean and keeping the team in sync does not need a dedicated role. We find the Product Manager role has absorbed this function, though rather than taking on coordination responsibilities, the “duties” of project manager grow naturally out of their traditional shaping of product direction.
Removing a layer takes with it the delays created by meetings and the time spent creating the reports and artifacts for those meetings. Velocity increases, but also your team spends more hours on the work they enjoy and fewer hours on the burdens of admin.
From laser to wave
Before agents, businesses had to be ruthless and laser focused on the right actions. Resources were limited. Options needed to be ruthlessly culled. Teams were a balancing act of size vs output.
Now, those same teams can be redistributed into 3x or 4x the pods. For some areas, like backend where the “product” might be pure code with no design or UX requirements, a pod could be a single developer driving agents.
Your product can advance on multiple fronts simultaneously, the laser becomes a wave of progress. Resources aren’t suddenly unlimited, but agents are clearly a multiplier.
At SoftwareSeni our extended teams are helping clients leverage that multiplier. If you’d like to chat about how a focused, agent-supported pod could help you compete in the current market, hit us up here.
