The bottleneck in enterprise modernization isn’t execution. It’s everything that happens before it.
Enterprise modernization is failing before it starts — not during implementation, but in the gap between understanding a system and being able to act on it. That is the central finding of new research from Sweep, an agentic platform for enterprise systems, drawn from more than 12,000 interactions across 500 operators at more than 200 organizations.
The numbers are striking. Across the dataset, 80 percent of the work performed by a typical Salesforce operator is dedicated not to building or changing systems, but to reconstructing an understanding of how those systems currently work. Every change, however routine, begins with an investigation.
The Cost of Context
The practical cost of that pattern is measurable and significant. Sweep estimates that administrators spend between 620 and 1,040 hours per year on system context reconstruction alone — translating to between $42,000 and $70,000 annually per administrator, or as much as $700,000 per year for a ten-person team.
The root cause is structural. Enterprise systems accumulate complexity over years of growth, staff turnover, and one-off projects. The three-stage lifecycle of enterprise system work — discovery, design, and build — is fragmented across tools, teams, and workflows, forcing each new project to restart the same investigative process from scratch.
Among the most active users in Sweep’s dataset, 80 percent of activity occurred in the discovery phase, with dependency tracing, automation discovery, and permissions analysis driving the majority of work. A further 19 percent occurred in the design phase. Execution — the work that actually appears on roadmaps and sprint metrics — accounted for a fraction of total effort.
Also Read: The AI Attack You Haven’t Heard Of. But Should.
AI Is Accelerating the Problem, Not Solving It
The emergence of AI tools has made it faster to generate flows, fields, automations, and code within enterprise systems. What it has not done is accelerate system understanding — and that asymmetry is creating a new category of technical debt.
Agentic tools that generate changes without a complete picture of system dependencies frequently introduce modifications that break other parts of the system. These failures often do not surface immediately, emerging days or weeks later when the original context has already been lost and the work of diagnosis must begin again.
Sweep’s research shows this pattern emerging clearly in early Agentforce deployments, where investigation rates climb sharply as systems mature and AI-generated metadata accumulates. Salesforce’s Headless 360 architecture, announced at TDX 2026, reflects a broader shift toward more programmable, agent-accessible systems — a shift that amplifies the problem if structured system understanding does not keep pace.
The Velocity Tax
The report introduces a concept it calls the Velocity Tax: the measurable cost of system work that must happen before execution can begin, and that rarely appears on any roadmap or sprint metric.
The human dimension of that tax surfaces in behavioral data. Planning activity more than doubled after 9 p.m., rising from 7.2 percent during the day to 15.7 percent at night — a signal that context reconstruction is overflowing standard working hours. Seven percent of all interactions in the dataset referenced legacy labels such as “DEPRECATED,” “DO NOT MODIFY,” or “DO NOT DELETE” — informal governance markers that emerge organically when systems become too complex to reason through directly.
“CIOs are fed up with modernization projects that drag on for years and cost millions,” said Ido Gaver, chief executive and co-founder of Sweep. “AI changes that. What used to take 12 months can now be completed end-to-end in days. The real issue is complexity. It kills velocity. Traditional system integrators have built businesses around that inefficiency. AI, paired with deep system context, removes that friction and restores speed.”
Also Read: Why AI Hallucinates When Identity Data Goes Stale
The Argument for Unification
The report’s prescription follows directly from its diagnosis. When discovery, design, and build are unified into a single continuous workflow — with system context maintained and compounded rather than reconstructed from scratch on each project — the investigative work that currently dominates administrator time can be compressed from days or weeks to minutes.
Without that unification, the Velocity Tax compounds with every new deployment, every new AI-generated change, and every new team member who must learn how the system works before they can touch it.
