How does this work with 300 people in non-tech corp? (Assuming the 1000 person org goes from 8 person team to 3 person teams.) what are the coordination mechanism? Communication? Today, regrettably, many orgs have a 10-30% tax on the tech team in the form of product manager, SME operations, go to market. That ratio would go to 30-60% without concomitant reduction. Managing that interface is more than a full time job.
Firstly, we're in experimentation mode, not scaling mode, so this is something to try for a tiger team, not to scale across the org today.
Secondly, communication and co-ordination will continue to be necessary, but should be mostly automated. If your PM agent understands the business context, is keeping a time series log of the backlog, knows who on the team is in and out and who is coding what, if they understand deadlines and dependencies (business and technical), if it has the notes from all of the conversations relating to the project, keeps an eye on all of the slack channels and is responsible for updating the wiki, documentation, user guides, project progress and all the rest, and all of that is projected onto a knowledge graph with deterministic checks to avoid hallucinations, if anyone wants to ask or know something, a quick chat with the PM agent should get them what they need.
You need to identify, decompose, re-envision and re-build agentic communications and workflows - this is a material reorienting, but for each communication interaction the first question is what is the purpose it solves, and the second is how can sufficient context be provided to a team of agents to answer the question and/or to bump it up to the appropriate human where real judgement is required that exceeds what can be done with the context embedded in the system.
Key implications:
- No private DM's in slack
- No emails unless the agentic PM gets cc'd
- No conversations that are not recorded - ever
It's the same hygiene you need for a remote team of humans, but it's particularly important otherwise you'll end up with agents making what would otherwise seem to be reasonable decisions based on the wrong information.
How does this work with 300 people in non-tech corp? (Assuming the 1000 person org goes from 8 person team to 3 person teams.) what are the coordination mechanism? Communication? Today, regrettably, many orgs have a 10-30% tax on the tech team in the form of product manager, SME operations, go to market. That ratio would go to 30-60% without concomitant reduction. Managing that interface is more than a full time job.
Firstly, we're in experimentation mode, not scaling mode, so this is something to try for a tiger team, not to scale across the org today.
Secondly, communication and co-ordination will continue to be necessary, but should be mostly automated. If your PM agent understands the business context, is keeping a time series log of the backlog, knows who on the team is in and out and who is coding what, if they understand deadlines and dependencies (business and technical), if it has the notes from all of the conversations relating to the project, keeps an eye on all of the slack channels and is responsible for updating the wiki, documentation, user guides, project progress and all the rest, and all of that is projected onto a knowledge graph with deterministic checks to avoid hallucinations, if anyone wants to ask or know something, a quick chat with the PM agent should get them what they need.
You need to identify, decompose, re-envision and re-build agentic communications and workflows - this is a material reorienting, but for each communication interaction the first question is what is the purpose it solves, and the second is how can sufficient context be provided to a team of agents to answer the question and/or to bump it up to the appropriate human where real judgement is required that exceeds what can be done with the context embedded in the system.
Key implications:
- No private DM's in slack
- No emails unless the agentic PM gets cc'd
- No conversations that are not recorded - ever
It's the same hygiene you need for a remote team of humans, but it's particularly important otherwise you'll end up with agents making what would otherwise seem to be reasonable decisions based on the wrong information.