Preparing people for AI-assisted work means building verification, judgment, agent-management and communication skills, with practice and management support.
From working with state government, I've heard many AI conversations focus on which tools are approved, what information can be used, and whether the output can be trusted. Those are necessary conversations. But they leave me asking a broader question: are people working in public service being prepared for the work that comes next?
A recent video from Maddy Zhang helped sharpen that question.
In her November 1, 2026 video, AI Writes My Code. What Am I Supposed to Do Now?, Zhang examines how AI is changing software engineering. Her argument centers on four skills: verification, judgment, managing agents, and communication.
Although she is speaking to engineers, I see a useful question for anyone introducing AI into an organization: what does an employee need to know once AI starts doing part of the work?
1. Verification: knowing what good work looks like
Zhang emphasizes the ability to inspect code someone else, or an AI agent, produced. Does it work? Where might it fail? What evidence supports accepting it?
For a public-sector team, that same discipline could apply to a generated summary, a proposed website change, or an analysis. Asking someone to review an output also means making clear what they should check and what evidence they need.
2. Judgment: deciding what deserves to be done
Her second skill is handling ambiguity: defining the problem, weighing alternatives, and making defensible decisions.
That matters before an AI tool starts generating anything. A faster answer to an unclear request can still leave the team with the wrong result. Employees need room to clarify the goal, question assumptions, and recognize when a decision requires someone else's expertise or authority.
3. Managing agents: designing how the work gets done
Zhang describes setting up AI workflows with clear specifications, review steps, and tests. An agent is an AI system that can carry out a sequence of actions using tools.
For an organization, that raises practical questions. What can the agent do independently? When must it stop? Who handles an exception? A second agent can help check the first, but agreement between them is not sufficient evidence that the work is correct.
4. Communication: making the necessary context available
Her fourth skill connects the other three. People and agents need clear requirements and accessible decisions. An important choice left in a private conversation may never reach whoever, or whatever, carries out the next step.
I take this as a reminder to make the purpose, constraints, expected result, and unresolved questions explicit.
What this could look like in practice
Imagine a fictional web team using AI to propose changes to a public information page. The assignment includes the source material, intended audience, and acceptance criteria. The output shows proposed edits, their supporting sources, and questions requiring a decision.
A content owner verifies accuracy, and an accessibility specialist evaluates relevant accessibility issues. The manager ensures the review has an owner and time allocated to it. That is a workflow the team could practice and assess.
The question for public service
California has already recognized the need for workforce preparation. Its published training guidance addresses employees, business leaders, and technical specialists. CDT's training offerings also include engineering and development, project management, and an introductory agentic AI course.
California is one example. The broader question applies across state, federal, and local government: how does that preparation reach everyday work?
In the conversations I encounter, hallucinations, meaning fabricated or incorrect AI output, remain a frequent concern. That concern deserves attention. So does preparing employees to recognize those problems, investigate them, and decide whether the work is usable.
As AI capabilities move quickly, are state workers, federal workers, and people across public service ready for this shift?
Before we expect employees to work alongside AI agents, are we giving them the training, practice, time, and support to verify the work, exercise judgment, direct the process, and communicate clearly?
And are their managers prepared to guide that work and recognize when a team is ready?
Sources and scope
- AI Writes My Code. What Am I Supposed to Do Now?
Maddy Zhang, November 1, 2026. Source of the four-skill framework. Reviewed using the full auto-generated English transcript; visual content was not assessed.
- California GenAI state staff training
Official guidance for workforce, business-leader and technical training, accessed November 1, 2026. Establishes the stated training approach, rather than measured workforce readiness.
- CDT Generative AI Training
Official technical training domains and course listings, accessed November 1, 2026. Course offerings can change.
Independent educational commentary. Source claims, interpretation and any observed results should be read separately. No employer endorsement is implied.