A Salesforce admin wakes up on a Tuesday morning to a fairly humdrum request: clean up an unused field.
Seems straightforward enough. The field hasn’t been updated in years. Leadership wants to simplify the CRM. Someone in RevOps swears nobody uses it anymore.
So the admin starts checking dependencies.
Huh. Well, a flow references the field. Then wait — another. One automation hasn’t fired in eighteen months, but nobody feels comfortable deleting it because nobody remembers why it existed in the first place. An integration points to the field through middleware documentation last updated during the Biden administration.
A Slack thread from 2022 suggests that the field might affect quoting logic for a single enterprise customer. Someone from Success vaguely recalls a consultant warning the team not to touch it.
Three days and hours of work later, the field is still hanging on.
A month from now, another admin will repeat the same investigation almost from scratch.
It’s a strange quirk about enterprise systems work that teams often don’t move linearly through their projects. They loop through them. The same investigations recur. The same questions surface. The same dependency tracing exercises repeat themselves across quarters, across teams, and sometimes across entirely different generations of employees.
In Groundhog Day, Bill Murray wakes up every morning to the same song and the same conversations, trapped inside a world that continuously resets itself.
For a lot of systems teams, enterprise operations can feel eerily similar.
Meaningful change begins with reconstruction
For years, conversations around enterprise productivity focused almost entirely on execution. How fast can teams ship? How quickly can they automate workflows? How much implementation work can AI accelerate?
But new patterns are emerging inside enterprise systems work, and they suggest something deeper.
According to our recent State of Enterprise Systems report, nearly 80% of operational effort happens before execution ever begins. Teams spend most of their time investigating, mapping, validating, and reconstructing understanding before they can safely make changes.
That changes the shape of the problem entirely.
The dominant workflow inside enterprise systems no longer looks like a straightforward pipeline from design to implementation. Increasingly, it looks more like a recursive investigation cycle: discover, rediscover, validate, investigate, design, verify, then finally build. The implementation itself may only consume a fraction of the total project timeline. The rest disappears into uncertainty reduction.
Enterprise systems often lose knowlege faster than they keep it
Admins leave for other jobs. Consultants roll off projects. Documentation drifts out of date. Naming conventions decay. Temporary workarounds harden into permanent architecture. Overlapping automations build up over time. Eventually, organizations inherit systems full of logic that still functions operationally but no longer survives conceptually. The machinery works, but the reasoning behind it has faded.
That’s why so much enterprise work feels archaeological. Systems teams spend enormous amounts of time reconstructing intent from fragments. A deprecated field becomes a clue.
A broken naming convention becomes a historical artifact.
An abandoned automation reveals the outline of a migration strategy nobody documented properly five years ago.
Most organizations think technical debt primarily slows execution. In reality, a huge percentage of technical debt slows comprehension.
And this is exactly why AI changes the conversation so dramatically.
The org has no clothes
For years, slow implementation speeds masked the reconstruction problem. If a migration took nine months, nobody paid much attention to the fact that three or four of those months disappeared into discovery work. The planning phase simply blended into the broader timeline.
AI compresses execution so aggressively that the real bottleneck has suddenly become visible. User stories generate instantly. Workflows scaffold themselves. Code suggestions arrive in seconds. Entire implementation layers accelerate. But system understanding does not accelerate at the same rate.
In many cases, it becomes even more important.
Faster generation creates more opportunities to modify systems, which increases the need for dependency awareness, governance, and operational certainty. The central enterprise question starts shifting away from “How fast can we build?” toward something more like “How confidently can we change what already exists?”
That helps explain why so many orgs feel both terrifically excited and terrifyingly uneasy about enterprise AI at the exact same time. The generation layer works insanely well. The problem sits underneath it.
Most enterprise systems contain enormous amounts of furtive relational complexity that AI alone cannot magically resolve.
Down the rabbit hole
One of the more revealing patterns in Sweep’s report involved “deep investigation sessions,” where operators asked ten or more consecutive questions while trying to understand a single system area.
Anyone who has worked in Salesforce recognizes this phenomenon immediately. A field cleanup turns into a flow audit. The flow audit exposes permission conflicts. The permission conflicts uncover undocumented integrations. Those integrations surface entirely different architecture concerns.
Every answer reveals another dependency. Modern enterprise systems contain layers of accumulated operational history that nobody fully sees at once.
That may be the real velocity tax inside enterprise organizations: repeated cognitive reconstruction. Teams don’t simply manage systems anymore. They repeatedly rediscover them.
Which is why context suddenly matters so much
AI agents don’t just need the ability to generate actions. They need operational understanding. They need metadata relationships, dependency awareness, permissions context, automation lineage, downstream impact analysis, and architectural memory. Otherwise acceleration simply increases the rate at which organizations create additional complexity for themselves.
The future advantage in enterprise AI may not belong to the organizations that build the fastest. It may belong to the organizations that preserve understanding the best.
For years, systems teams accepted reconstruction work as an unavoidable part of enterprise operations. Open the ticket. Trace the dependencies. Ask around in Slack. Rebuild the context. Start over.
But AI changes the visibility of the workflow. When execution compresses from months into hours, organizations can finally see where the real work lived all along — deep in there, deep in understanding.
