AI automation is stalling on missing context, not poor technology, according to Latch’s State of AI Automation at Work 2026, a survey of 742 U.S. working professionals released today, along with Latch’s new accounting automation platform.
Among respondents whose AI project fell short, 57% blamed missing context, such as bad data or a system that didn’t understand how the work is done. Only 8% said the technology wasn’t capable enough.
Other findings:
- AI still needs a babysitter. 73% say AI output needs a person to intervene at least sometimes, rising to 84% among finance and accounting professionals.
- Most work isn’t written down. Respondents estimate only 41% of recurring work is documented well enough to hand off.
- Exceptions are the rule. 83% say their regular work involves judgment calls written instructions don’t cover.
- Know-how walks out the door. 61% say their organization rehired a former employee for their institutional knowledge.
“AI isn’t missing intelligence; it’s missing what your best accountants know, such as which variance to ignore and which vendor always needs a second approval,” said Stefan Kalb, Cofounder and CEO, Latch. “Latch lets accountants teach that once, hand off the repetitive steps, and focus on the judgment and advisory work only they can do. Their expertise doesn’t get replaced. It gets amplified.”
Unlike most AI tools, which at most capture screenshots or record video to get a deeper understanding of work, Latch actually asks questions, eliciting information about what happens when processes differ from written procedure. Employees just show Latch a task while Latch watches and asks questions, building institutional context about what happens when the process isn’t clear. Latch then builds a “sidekick” that automates the work to save time, reduce errors, and turn the focus to the interpretation of the results.
Latch’s own finance team turned a month-end Xero-to-Google Sheets workflow into a scheduled sidekick, cutting more than 200 clicks to about a dozen and saving 15 minutes of manual work per run.
“AI can be a game changer to automate financial and accounting workflows,” said Kalb. “But securely implementing AI across your team can be difficult and costly, especially when that process involves significant investment from your IT team. Latch understands natural language and video intuitively, asking questions the same way a human teammate would, and uses that as the basis of building useful AI automations.”
Accounting and finance teams that email Kalb at stefan@latch.ai by Nov. 30, 2026, will receive 1,000 free Latch credits.
Latch turns how people actually do their work into automations, without a single prompt. Show Latch a task and talk it through; it captures the steps, exceptions and reasoning, then builds sidekicks that run the work the same way every time and stop to ask when a decision needs a person. That knowledge is also available to any AI agent over MCP. Founded by Stefan Kalb and Jared Kofron, Latch is based in Seattle and backed by Fuse, Soma Capital, Garry Tan, Liquid 2 Ventures and Pioneer Fund. Learn more at latch.ai.
Sign in to get access to this free resource, and all of our whitepapers and reports.
Download this content today!
Register Now Already registered? Click here to Log In
