Leading in the AI Era Module 1 · Lesson 1.2 · What these systems can and cannot do for your organization

Four places it lands

Updated

Beginner

What you'll learn

~11 min
  • Name the old constraint, the new capability and the new feasibility in four kinds of work
  • Name the human verification that each kind of work still requires
  • Compare these examples with one workflow in your own organization

You leave this lesson with: one workflow of your own named, with its old constraint and the person who must verify the result.

These four examples show the systems at work and suggest workflows worth examining. The systems are AI tools that work from plain-language instructions. The examples are illustrations, not measured results or promises about your organization.

Scientific research: a university research group

A university research group studies protein samples. Analysts write custom scripts for each new instrument export, and a graduate student spends weeks on each one.

  • Old constraint. Only people who could program could turn raw instrument output into an analysis.
  • New capability. A researcher describes the analysis in plain language, and the system drafts the script and a first set of charts.
  • Newly feasible. Researchers use drafted scripts to try more ways of analyzing the same samples before consulting the statistician.
  • Verification still required. A person who knows the method checks the code against a sample with a known answer, and the statistician approves the analysis before it goes in a paper.

Mission operations: a government agency

A government agency’s operations team reviews incoming reports from field offices every morning. Staff read each report and copy facts into a tracking system by hand.

  • Old constraint. Reading volume limited how many reports could be reviewed on the day they arrived.
  • New capability. The system reads each report, lists the facts it found with the sentence each came from, and flags reports that look unusual.
  • Newly feasible. The team uses assisted extraction to review more reports on the day they arrive.
  • Verification still required. A staff member confirms each flagged fact against the report before it enters the record, and a supervisor samples unflagged reports to see what the system missed.

Application modernization: a company

A company runs an inventory application written many years ago. Few current staff understand how it works, and replacing it has been repeatedly postponed.

  • Old constraint. Nobody could afford the weeks of reading old code needed to estimate what replacement involves.
  • New capability. The system reads the old code, describes in plain language what it does, and lists the rules buried in it.
  • Newly feasible. A business owner and an engineer can review that description together and use it for an estimate and a first working version to compare with current practice.
  • Verification still required. The estimate is only as good as the list of rules, so an engineer checks a sample of the old code for rules the description missed. The business owner confirms each rule is still wanted, engineers test the new version against the old one, and security and operations review before it runs for real users. A first version that works in a demonstration is not yet ready for staff to rely on every day.

Institutional knowledge: a public agency’s policy archive

A public agency holds decades of policies, memos and prior determinations in its policy archive, spread across shared drives. New staff ask the same questions of the same few long-serving colleagues.

  • Old constraint. Knowledge lived in people and in archive files that no search could answer questions about.
  • New capability. The system answers a question from the agency’s own documents and points to the passage it used.
  • Newly feasible. New staff get a first answer without waiting for a colleague, leaving long-serving staff more time for judgment calls.
  • Verification still required. A person checks that the cited passage supports the answer and that the document is current. Lesson 1.3 explains why a cited answer is not yet a checked one.
💬Your turn, one workflow

Choose a workflow in your area. Which of the four patterns does it resemble? Who knows what a correct result looks like, and could that person verify one result this week?

Bring the candidate workflow to the person who currently checks its results. Ask whether the proposed assistance addresses the actual constraint or merely moves work into checking and correction. Record an alternative that does not use AI, such as clearer source material or a simpler request form. The examples do not establish that any particular workflow needs a new tool.

Key takeaway

LEADERSHIP DECISION
Choose one workflow whose main constraint is reading, writing or searching, and name who will verify its results.
IMMEDIATE ACTION
Ask that person to describe, in one paragraph, what a correct result looks like and how they would check one.
SUCCESS MEASURE
Calendar days for one real case to go from request to verified result, with checking and correction time no higher than before.
Search lessons