The ceiling most AI investments hit
When a company invests in AI, the usual expectation is cost savings by doing the same things with fewer people. But that framing may be more optimistic than what is actually happening. AI is an added cost, and most companies are not reducing headcount, because more output still needs a human to review it, act on it, and take responsibility for it. The result, for many companies, is higher costs and more work, not less. The AI is producing things. Whether those things are moving the company toward its goals is a different question, and usually an unanswered one.
The question worth asking is not "is it working?" but "what is it actually doing for the company?" There are three real outcomes: AI adds cost with no clear gain, which is the most common scenario. AI reduces cost by genuinely replacing work, which is happening but usually in fewer instances than needed to impact the bottom line. Or AI enables the company to do things it could not do before, in the direction it is actually trying to go, and this is the only outcome that grows the company. All three can show up in a status update as "AI is working," but only the last one moves AI from a cost to a competitive advantage.
A company can only move as fast as its slowest part. If AI speeds up the slowest part, the company can do more. If it speeds up something else, costs may fall, but the company does not actually do more.
In many companies, the slow part is not the doing. It is deciding what to do, getting people to agree, and fitting different people's work together. That is the coordination layer, and it is where most organizations experience their real constraint, regardless of how much AI they have deployed on the execution side.
Personal agents take you only so far
Personal AI agents are good at doing work once the task is clear: drafting, calculating, summarizing, looking things up. They are less good at the work that comes before: deciding what is worth working on, agreeing on what counts as a good answer, noticing the questions that have not been asked.
Each person produces more output, but that output still has to be reviewed, prioritized, and reconciled with everyone else's. That coordination work does not go away, and there is just more of it to handle. And because someone still has to be in the loop for every piece of output, companies are not reducing headcount in a way that makes an impact. They add an AI cost on top of the headcount they already have. The company is not doing something new; it is producing more of the same, at higher cost, with more to manage.
Specialized workflow agents hit the same ceiling
Specialized agents handle a single workflow: processing invoices, triaging tickets, scoring leads. The efficiency gain is real, but it has a ceiling set by what the workflow used to cost. And these agents are not free to run: they need monitoring, maintenance, and someone to step in when something goes wrong. The work does not disappear; it just shifts from doing the workflow to watching over it.
The deeper limit is the same. A workflow agent improves how an existing process runs. It does not ask whether the process should exist, or whether the company should be solving a different problem instead. The result is incremental efficiency on work the company already does, and no path from that efficiency to new capability, or to the goals the company is actually trying to reach.
What we are investigating
City by the Sea studies the gap between what AI can produce and what organizations actually need to move forward. We believe the bottleneck in most knowledge-work environments is not execution capacity but decision throughput: the rate at which an organization can identify what needs to be decided, get the right people aligned, and turn that alignment into coordinated action.
Most current approaches to AI deployment treat the organization as a collection of individual workflows to be optimized in isolation. We are interested in what happens when you treat the organization as a coordination system, and ask how AI can improve the quality and speed of the decisions that hold everything together.