When AI speeds up production, the constraint shifts to review, decision rights and the physical spaces where people interpret its work.
Anthropic recently described an unusual internal problem relevant to AI workplace design.
As the company's use of AI coding increased, the number of merged code changes per day rose significantly, from approximately 2,500 in May to over 30,000 by September. This rapid growth made Git a bottleneck, prompting Anthropic to develop and transition to a new source control system called Scion. Nevertheless, human engineers still review every change request. Their focus has shifted from examining every line of code to assessing the intent behind the changes and rejecting any requests that violate core constraints or respond excessively to hypothetical risks.
This is not a benchmark for the average company. Anthropic is explicit that its workload and internal token spending are unusual. It is still a useful stress test.
When AI makes one part of work dramatically faster, the bottleneck does not disappear. It migrates into the surrounding system.
That system includes data, integration, review capacity, decision rights, incentives and, increasingly, the physical environment where people interpret and act on AI-generated work.
What the research does and does not say
The latest enterprise research makes the same point, although with more nuance than the usual "redesign your workflows" advice.
ISG's 2026 survey of 400 senior decision-makers found that most business and financial outcomes were falling short of expectations even as adoption, model performance and governance were rated more positively. The reported constraints included data quality, technology integration, architecture complexity, operating cost, and the difficulty of scaling human review. Workflow design was one factor among several, not the sole explanation.
That distinction matters because weak returns may reflect immature technology, measurement lag, or a use case that never deserved investment.
Some organizations are still learning what the technology can reliably do. Others are trying to layer it onto fragmented systems and inconsistent data. Not every disappointing result is a failure of organizational imagination.
But once the technology begins to work, organizational design becomes harder to avoid.
If AI can produce drafts faster than people can evaluate them,
the problem becomes review capacity.
If it can summarize meetings faster than teams can resolve disagreement,
the problem becomes decision discipline.
If it can coordinate tasks across systems,
problems shift to who's authorized to commit the organization.
And if AI begins to operate through microphones, cameras, displays and sensors embedded in rooms, the workspace itself becomes part of the operating model.
When the room participates in the work
That last point is easy to dismiss as a furniture or facilities question. I think it is more consequential.
A room that can identify speakers, capture discussion, retrieve records and generate follow-up is no longer a passive container for work. It is participating in the workflow.
That changes the questions we need to ask.
When is the room listening and how do people know? Can someone turn the system off without finding an administrator?
Where can a team place an AI-generated recommendation beside the underlying evidence and examine both?
Where can a person work privately with an AI system before bringing an idea to the group?
AI Workplace Design: The evidence is still early
A smart-office study, which included 24 interviews, revealed that privacy preferences varied depending on the type of data being collected and the physical environment. It also showed that seemingly harmless environmental sensors could lead to inferences about occupancy, activities, and routines that users might not fully understand.
Additionally, a 2026 workplace-monitoring case study found that employees were generally open to monitoring for educational purposes. However, they still had concerns regarding the accuracy and use of the data generated. The involvement of trusted managers who could interpret and question the system's outputs was found to be very important.
While these findings are valuable, they should not be overly generalized, as both studies were qualitative with limited sample sizes. Nevertheless, they highlight a design challenge that cannot be resolved by software policies alone.
What an AI-aware workplace may need
An AI-aware workplace may need:
Visible capture states
Meaningful off controls
Spaces for private human-AI interaction
Shared surfaces for verification
Acoustics that protect both intelligibility and confidentiality
Clear transitions between machine-supported exploration and human decision
It may also need less technology than vendors assume.
The strongest counter-case
The strongest counter-case is that most organizations would gain more from cleaning their data, simplifying their processes, and clarifying ownership than from embedding AI into rooms. I agree. A sensor-rich workplace layered on top of incoherent work would create a more expensive form of incoherence.
The point is not that every office needs to become intelligent. It is that wherever AI enters the environment, the environment begins to shape who notices, who decides, who is interrupted, and what becomes official.
We have spent years treating workplace design and digital workflow design as separate disciplines. Ambient AI makes that separation harder to sustain.
About this article
This article reflects the analysis and opinions of Signal & Context Advisory as of its publication date.
It is provided for general information and discussion and is not legal, regulatory, financial, security, or other professional advice. Decisions about your organization should rest on advice suited to its specific circumstances.
Examples are illustrative and do not describe any client unless stated. Where external sources are used, they are linked in the text. Mentions of third-party companies, products, or trademarks are for commentary only and do not imply affiliation or endorsement.
Use of this site is governed by our Terms and Privacy Policy.
© 2026 Signal & Context Advisory LLC. All rights reserved.