Saturday Tools: Found So Doing

Each Saturday, The Wise Operator digs into the tools, tutorials, and trending builds worth your weekend. No news, just hands-on discovery.

It’s Saturday. The news can wait. Three releases this week land in the same direction. Z.ai opened a 753 billion parameter model purpose-built for coding work that runs for hours without an operator at the keyboard. OpenAI graduated scheduled tasks out of beta and gave a billion-user chatbot a built-in cron. Google sunset its old Gemini CLI and forced every terminal user onto Antigravity, a Go binary with subagents, hooks, and async background workflows. The unit of work in 2026 stops being the prompt. It becomes the standing instruction.

Tools Worth Your Weekend

GLM-5.2 from Z.ai

Z.ai dropped the full MIT-licensed weights of GLM-5.2 on Hugging Face and ModelScope on June 17. The shape matters before the score. 753 billion total parameters in a mixture-of-experts layout, a new IndexShare sparse-attention design, and a usable one million token context window. The benchmarks underneath: 62.1 on SWE-bench Pro and 74.4 on FrontierSWE, both ahead of GPT-5.5 and within range of Opus 4.8, at roughly one-sixth the per-token cost on Z.ai’s hosted API.

Why it matters: an open-weight model that actually wins on the long-horizon coding benchmarks is a budget event, not just a leaderboard event. The agent economy until now metered every token a background agent read or wrote. A self-hostable model trained for multi-hour autonomous coding work is the first credible exit from that meter. The China-data-risk headline is real for hosted API calls and irrelevant the moment you download the weights, which is the actual point of an MIT license.

Try this: Do not try to swap out your daily driver. Pick a chore you have been paying tokens to avoid, a repo-wide audit, a dependency upgrade across three projects, a refactor that would touch nine hundred files. Spin up GLM-5.2 on Z.ai’s hosted endpoint for one Saturday afternoon and let it run unattended. Compare the diff to whatever your current setup would have produced and compare the bill. The point is not to switch this weekend. The point is to learn where the meter stops making sense for the unattended work, because that is the work that is about to grow. Free open weights (MIT). Paid API at roughly one-sixth the cost of GPT-5.5.

ChatGPT Scheduled Tasks

On June 17, OpenAI moved scheduled tasks out of limited beta and into broad rollout on Web, iOS, Android, and macOS. You write a prompt once, attach a one-off or recurring trigger, and ChatGPT runs the prompt on its own clock and pushes you the result. A dedicated Scheduled page lets you pause, edit, or delete the standing job. Plus, Pro, Team, and Enterprise. Ten active tasks per seat.

Why it matters: the world’s most-used chatbot just grew an agency the chat tab never had. Until this Saturday, every ChatGPT interaction started with a human pressing send. Starting this Saturday, ChatGPT can start the conversation. That is a small product change and a large ambient agent change. The seat that was a search-bar replacement becomes a standing instruction the operator no longer thinks about, until it pings him.

Try this: Do not start with calendar triage. Pick one piece of weekly intelligence work you have been outsourcing to a paid SaaS, a competitor-price scrape, a job-board scan, a Reddit listen for your category. Wire it as one scheduled task that runs Monday at seven, checks the ten URLs you actually care about with web search, and emails you only the diffs that mention price or a new product name. Run it for three weeks before you trust it. The exercise is not to cancel the SaaS. It is to feel what the operator role becomes when the work starts itself. Paid only, included on Plus and up.

Antigravity CLI from Google

On June 18, Google’s Gemini CLI stopped serving requests for AI Pro, Ultra, and free-tier users. Antigravity CLI is now the only supported terminal agent for the Gemini 3.5 family. Built in Go, not Node. Async background workflows, subagents, hooks, and a plugin system that mirrors the same harness Google uses internally. The forced migration broke a lot of overnight automation. The destination is the first apples-to-apples terminal competitor Claude Code has had.

Why it matters: terminal agents had been a one-vendor conversation for two years. Antigravity ends that. Two viable harnesses with subagents, hooks, and skills is a different market than one. For the operator who keeps a long-running build going in a tmux pane, the question stops being which lab makes the smartest model and starts being which CLI runs the most useful parallel work while you are doing something else. Free with any Google AI account.

Try this: Do not swap your daily CLI. Stand up Antigravity in a fresh worktree this weekend and give it the one job your current setup cannot do well: a parallel audit across all your active repos, one subagent per repo, each running the same security and drift checklist in the background while the parent agent aggregates findings into one report. The output is interesting. The reason for the exercise is the harness shape, because the operator who has only ever run one CLI has only ever seen one shape, and that is the shape his hand reaches for next. Free with Google AI Pro or Ultra.

What to Watch

Claude Code + YouTube = $62,000/Month (Full AI Tutorial) is the rare monetization video that doubles as a real Claude Code workshop. Skip past the headline number. The useful segment is the back half, where the creator walks through the slash commands, the file-based prompt orchestration, and the tool-chain that drives a multi-step content pipeline end to end. The mechanic transfers. Substitute “faceless YouTube channel” with “newsletter” or “client deliverable” or “course generator” and the same wiring works. Watch it before you spin up another standing instruction this week. The lesson is what an orchestration looks like when one person built it for himself.

Wisdom Speaks

“Do everything as if Epicurus were watching you. It is, indeed, nobler by far to live as you would live under the eyes of some good man, always at your side; but nevertheless I am content if you only act, in whatever you do, as you would act if anyone at all were looking on; because solitude prompts us to all kinds of evil.” Seneca, Moral Letters to Lucilius, Letter 25.5

Seneca borrowed Epicurus to name a problem the Stoics were not embarrassed to share with their rivals. An unsupervised servant in unwatched solitude is a danger to himself and to the household. The Stoic answer was to install an imagined watcher inside the servant, so that conduct under no eye still tracked conduct under one. This week’s tools turn the problem inside out. GLM-5.2 runs for six unattended hours on the operator’s behalf. A ChatGPT scheduled task fires at three in the morning while the operator sleeps. An Antigravity subagent forks at dinnertime and finishes before the dishes are dry. The servant is now literally in solitude. The operator’s stewardship question is no longer “what should I prompt.” It is what should be found being done when I walk back into the room.

The biblical landing is Matthew 24, the parable of the faithful and wise servant. “Who then is a faithful and wise servant, whom his lord hath made ruler over his household, to give them meat in due season? Blessed is that servant, whom his lord when he cometh shall find so doing” (Matthew 24:45-46, KJV). Christ does not bless the servant who waits for the master. He blesses the servant who is found so doing. The wise master goes away, gives the standing instruction, and the wise servant is still at it when the master returns. The folly is not in the absence. The folly is in the delay-loophole. “My lord delayeth his coming” is the inner sentence of the foolish servant, and it is the inner sentence the operator is one scheduled task away from. The discipline this Saturday names is epimeleia, the diligent care of an entrusted charge in the master’s absence. Every standing instruction you write is a small attempt to encode it. Every guardrail is the eyes of some good man, always at his side. Write them like the master is coming back today, because in the only meaningful sense, he is.

Last Saturday in this seat: Saturday Tools: The Apprentice Returns, on the week the workshop turned back to the operator and asked what the agent was making of him. This Saturday the workshop empties out, and the question is what the agent is making of the work while no one is watching.