How to automate your daily work with free CLI agents

TLDR: The free agents’ quiet revolution is not in coding: it is in automating. Gemini CLI and Qwen Code are invocable from any script with daily limits that comfortably cover personal use, and Goose chains flows that cross tools. The errands you did by hand every week (cleaning, summarizing, checking, transforming) become commands that run alone. Total cost: zero.

This guide is for developers and technical profiles with recurring errands: the fifteen-minute tasks repeating weekly that never justified automation because writing the script cost more than doing them. Free agents change that math: the script is now described.

It is also the natural complement to the strategy in the cost of coding with AI guide: the free agent as the volume workhorse.

1. Inventory what you do by hand every week

Profitable automation starts with the inventory, not the tool. A week noting the repeated errands produces the real list: the report you assemble every Monday from three sources, the cleanup of branches and temporary files, the check that environments are still healthy, the log summary when something complains, the data transformation that arrives in one format and is needed in another.

Good candidates share three traits: they repeat (frequency pays for the automation), they follow a pattern (even with variations that used to demand human judgment, which is exactly what the agent adds over the classic script) and they are verifiable (you can check they came out right).

Sort by frequency times duration: the twenty-minute weekly errand is worth more than the hour-long monthly one.

2. Build the base with Gemini CLI or Qwen Code

The central piece is an agent invocable from the command line, and the free tiers give it away: Gemini CLI with its broad daily allowance and long-context backing, and Qwen Code with thousands of daily requests that for personal automation are unlimited in practice. Both open source, both able to take an instruction and execute it with tools.

The usage pattern is the described errand: where you used to write a fragile fifty-line script, the script is now three lines invoking the agent with the instruction (“read today’s logs, summarize the errors new since yesterday and write the summary to this file”) and the agent supplies the flexibility the classic script lacked: the format shifted slightly, the date arrives differently, and it does not break.

Keeping both configured takes minutes and gives redundancy: if one hits limits or fumbles a task, the other steps in.

3. Automate cross-tool flows with Goose

When the errand does not live in the terminal alone (it touches the browser, an API, a service), the CLI agent runs short and Goose enters: Block’s open agent whose extension system connects to your flow’s tools, built exactly for chaining tasks across systems. Free, runs locally and works with the model you choose, including the free ones from this same flow.

The task type where it shines: the three-system flow you did by hand with six tabs open. Configuring its extensions is an afternoon of technical work, and every automated flow afterwards is a natural-language commission.

The complete stack settles into layers: CLI agents for terminal errands, Goose for the crossing flows, and your paid agent (if you have one) reserved for judgment work, as the two-agent strategy dictates.

4. Schedule, verify and maintain the system

Real automation runs alone: the system’s scheduled tasks (cron or equivalent) invoke the agent at its hour, and the result appears done. That jump demands the piece hobbyists skip: verification.

Every automation carries its check: the result exists, has the expected shape, and if something is off, a warning instead of silence. With agents, verification can be intelligent (“check this summary mentions every error in the original log”) but the rule is classic: automation without verification is a scheduled breakdown.

Maintenance is light but real: reviewing the warnings, adjusting instructions when the environment changes and pruning automations no longer used. The mature system is a silent employee doing the week’s errands, and building it cost exactly zero in tools. The whole sector lives in AI for software development.

Common mistakes

Automating what you do twice a year. The return lives in frequency: the repetition-sorted inventory prevents the hobby of automating for sport.

Scheduling without verification. Automation that fails silently is worse than manual work: the failure surfaces weeks later, damage accumulated.

Giving the agent more permissions than the errand needs. A scheduled agent with broad access is unnecessary risk: each automation with exactly its task’s permissions.

Ignoring the daily limits. Frequent automations add up requests: split across the two free agents and schedule the heavy ones at different hours to stay under the ceiling.

Frequently asked questions

Is all this really free?

The tools yes: Gemini CLI, Qwen Code and Goose are free, and the free tiers’ daily limits cover personal automation comfortably. The real cost is the afternoon of setting it up.

What happens when the agent fumbles an errand?

That is what verification is for: the warning arrives and the errand gets done by hand that day, as always. The difference from the classic script is that the agent fails less on small variations, which was the usual cause of breakage.

Can I use this at work with company data?

With your company’s data policy in hand: free agents send content to their providers. For sensitive data, Goose with a local model keeps everything at home, in exchange for the capacity of whatever model you can run.

Where do I start this afternoon?

Install Gemini CLI or Qwen Code, pick your inventory’s most repeated weekly errand and turn it into a three-line script with its verification. One automated errand running alone teaches the whole pattern.