Digital tools are becoming an increasingly important part of deep foundation construction. Estimating, scheduling, submittals, inspection notes, equipment records, progress reports, pile logs and project documentation now generate a dense digital trail.
As AI becomes embedded in more of these workflows, contractors and project teams are gaining new ways to process information and support decisions. But there is a risk that comes with greater automation: losing the context behind the data.
Deep foundation work takes place where drawings, models, soil conditions, equipment behaviour, contractual requirements and field judgment intersect. An AI-assisted output can appear reasonable and still be wrong because it relies on an outdated drawing, overlooks a site constraint, misinterprets an observation or treats incomplete information as complete. For that reason, one simple control deserves more attention before AI-assisted workflows become routine: the field exception log.
When the field challenges the software
The concept is straightforward. When a worker, foreperson, engineer, inspector, estimator or project manager questions or overrides an AI-assisted output that could affect safety, quality, cost, schedule, scope or contractual compliance, the project records the exception. The purpose is not to document every AI prompt or minor software error. Instead, the log captures the moments when human judgment materially changes what the system recommends.
A practical field exception log could record eight key pieces of information:
- Project and location
- AI-assisted task or tool
- Source material or evidence considered
- Responsible professional or field lead
- Exception or challenge
- Correction or override
- Downstream action
- Closeout lesson
These fields do not necessarily require another standalone system. They could be incorporated into an existing quality, change-control, non-conformance, daily-report or lessons-learned workflow. The important point is to preserve not only what went wrong but why a person challenged the output and what happened as a result.
Small errors can have big consequences
AI does not need to experience a dramatic failure to create problems on a foundation project. Small context errors can be enough. Consider a few scenarios.
An AI document assistant summarizes a specification but leaves out a requirement contained in a newer revision. A scheduling tool recommends a sequence that conflicts with site access or equipment availability. A progress-reporting system combines measurements from different locations. An image analysis tool flags a harmless condition that an experienced inspector recognizes immediately or fails to identify a condition the inspector considers significant. A retrieval assistant could also cite a superseded drawing, giving the impression that its recommendation is based on current project information when it is not.
None of these situations necessarily involves a sophisticated technical failure. They are failures of context. That distinction matters in deep foundation construction because conditions can vary significantly across relatively short distances. A bore log, pile driving record, test result, site observation or equipment reading can change an assumption that appeared reasonable in an office model.
Turning corrections into knowledge
A field exception log can turn those individual corrections into organizational memory. Without a record, a project team may resolve an immediate problem and move on without preserving the reason for the decision. The next crew, office team, subcontractor or project could then encounter the same issue and repeat the investigation. The exception log creates a record of where digital assistance fell short and where professional judgment changed the outcome.
If a crew member believes an AI-assisted recommendation conflicts with field conditions, they should not have to prove that the software is defective before raising the concern.
This aligns with principles outlined in the Project Management Institute’s Standard for Artificial Intelligence in Portfolio, Program and Project Management, which emphasizes governance, data quality, human oversight, escalation and responsible use.
For deep foundations contractors and their engineering partners, an exception log provides a practical way to apply those principles in the field. It answers a straightforward question: Where did digital assistance require professional correction, and what changed afterward?
Keep human judgment in the workflow
The log is only useful if workers have clear authority to challenge an AI-assisted recommendation. Companies should establish explicit stop-and-challenge procedures that identify who can pause an action, who reviews the evidence and who has final decision-making authority. If a crew member believes an AI-assisted recommendation conflicts with field conditions, they should not have to prove that the software is defective before raising the concern.
They should only need to demonstrate that the potential consequence warrants human review. That distinction can help create a culture where questioning an automated recommendation is treated as part of responsible digital practice rather than resistance to new technology. The objective is not to make workers distrust AI. It is to make sure they understand when professional judgment must take priority.
What to demand from vendors
The National Institute of Standards and Technology’s AI Risk Management Framework stresses governance, measurement and management across the technology lifecycle. That same logic can strengthen procurement decisions. Buyers of AI-enabled project software should ask whether a tool:
- Preserves source references
- Distinguishes current documents from superseded versions
- Records user overrides
- Maintains role-based permissions
- Exports histories that remain accessible when staff or vendors change
The exception log can also provide evidence during vendor evaluations. If a particular tool repeatedly produces the same type of error, that pattern can inform future purchasing decisions and help contractors identify where additional controls are needed.
From exceptions to improvement
The goal of an exception log is not to create a large database of AI mistakes. It is to identify recurring failure points and use that information to improve the way people and technology work together.
A contractor might discover that automated summaries repeatedly miss revision notes. A project team could find that scheduling recommendations break down under specific access constraints. An engineering group might learn that generated reports require a mandatory source check before release. Those patterns provide evidence that can guide workflow changes, staff training and vendor requirements.
A company could start with one AI-assisted workflow and run the exception log for 30 days. Teams could record only consequential exceptions, review them with field and office staff and then make one or two changes based on what the evidence shows. If the same exception stops occurring, the organization has learned something measurable.
Digital foundations still need field knowledge
The promise of digital foundations is not simply the addition of connected software to foundation construction. The real opportunity is to build digital systems that make information more useful without disconnecting it from the people who understand the physical conditions behind that information. A field exception log is a small operational change that can help make that principle tangible. As AI-assisted workflows become more common, recording where human judgment changes the digital answer could become one of the most valuable sources of information a foundation contractor has.
Digital foundations should not mean replacing field knowledge with software. It should mean building digital systems that pre serve engineering judgment, field experience and accountability as automation scales.
Gleb Tsipursky, PhD, is a behavioural scientist, CEO of Disaster Avoidance Experts and author of The Psychology of AI Adoption at Work: From Resistance to Results (Georgetown University Press, 2026).