Skip to content
explainroo
Working with AI agents

Agentic Coding: How Developers Work with AI Agents

Updated 4 min read

With explainroo How to Prompt a Coding Agent to Make an Explainer Video
On this page
  1. What developers delegate to coding agents
  2. How to keep agent work reviewable
  3. Where human review matters most
  4. Limit the agent’s access

When developers use an AI agent to change software, they give it a goal and access to tools such as a code editor, terminal, or test runner. The agent can inspect files, propose a plan, edit code, and run commands. The developer sets the boundaries, checks the results, and decides what ships. This way of working is called agentic coding.

The main change is that developers delegate a defined task instead of asking for a code snippet. Developers still make the product and engineering decisions; the agent handles steps within the task. Adoption is growing, but trust remains limited. The useful approach is to delegate work that can be checked, then verify the code and control what the agent can access.

What developers delegate to coding agents

Developers use agents for tasks with several linked steps, such as tracing a bug, changing code, and running tests. A JetBrains survey conducted in May–July 2026 found that 90% of professional developers used coding agents at work at least weekly, and 68% used them daily. Those figures apply to the survey’s respondents and timeframe, not to every developer.

A coding agent can save time on a clearly defined task when the developer can inspect its work. For example, a developer might ask an agent to trace a failing test, identify the relevant code, make a focused change, and report which tests it ran. The developer can then review the changes and decide whether the fix matches the intended behavior.

The human remains responsible for the goal and the quality of the result. Developers set requirements and resolve design questions; they also judge whether test coverage is sufficient. A practical workflow described by developer Tim Deschryver uses repository instructions and task-specific guidance to help agents follow project conventions. A file such as AGENTS.md can explain how to run tests or organize changes. Focused skill files can give extra guidance for specific tasks.

How to keep agent work reviewable

Start with a task small enough to verify. Describe the expected behavior and relevant constraints, then explain how to check the result. If the task touches unfamiliar code, ask the agent to inspect and explain its proposed change before editing. For a focused bug fix, that might mean confirming the failing behavior, changing the narrowest relevant code, and running the related tests.

Review the actual changes, not just the agent’s summary. Read the diff, check that the tests exercise the intended behavior, and run additional checks when the change affects other code. An agent can produce code that looks plausible while missing a requirement or creating extra work during debugging. In the 2025 Stack Overflow Developer Survey, 66% of respondents named nearly correct answers as a frustration, and 45% said debugging AI-generated code took more time. Those findings help explain why a passing test and a human review matter more than a confident-sounding explanation.

Provide project context in a form the agent can use. A short instruction file can point to the relevant directories, explain local conventions, and name the checks developers expect. Keep task-specific guidance focused so the agent sees what matters for the current work. If the agent repeatedly misunderstands a requirement, improve the instructions or narrow the task instead of letting it make a series of guesses.

Where human review matters most

Agent use does not guarantee faster work. A task takes longer if correcting the agent takes more time than the agent saves by writing code. A randomized trial summarized by OS-SCI found that early-2025 AI tools slowed experienced open-source developers by 19%. That result applies to a specific study, time period, and group of developers. It does not show how every agent performs today. It does show why developers should judge a tool on their own work rather than assume that delegation automatically increases productivity.

The 2025 Stack Overflow survey also found that 46% of respondents distrusted AI tool accuracy, compared with 33% who trusted it. Developers were especially reluctant to use AI for deployment and monitoring: 76% said they did not plan to use it for those tasks. Give the agent less freedom when mistakes would be costly. Let an agent draft or test a change with limited access; keep a person responsible for decisions that affect production, sensitive data, or users.

Limit the agent’s access

An agent that can read files and run commands acts through the permissions it receives. Each connected tool gives the agent more capabilities. Broad credentials and unrestricted access give mistakes or malicious instructions more room to cause harm. Security guidance from Checkmarx on AI agent risks identifies threats including prompt injection and data leakage. It also flags unsafe tool use and excessive permissions.

Give the agent only the access needed for its task. Scope credentials, validate inputs and outputs, and monitor tool use. Require human approval for high-risk actions. Treat instructions found in external content with care: text in a web page or repository can attempt to redirect an agent’s behavior. An agent should not receive authority to deploy, expose secrets, or make consequential changes simply because those actions are technically available.

Free and open source

Let your AI agent make the video

explainroo lets a coding agent like Claude Code or Codex make narrated explainer videos and product demos. The voice, the word timing and the rendering run on your own computer, with no API key and no cost per video. Optional outside services use your own key.

Copy this into your AI agent

Make me a short explainer video about [your topic]. Use explainroo for it: run npx --yes explainroo docs and follow what it prints.

Getting started Example videos GitHub

Made with explainroo: How explainroo makes a video

More on working with ai agents

All guides on working with ai agents