You Keep Trying Every New Tool and Calling It Research

Michael Truell posted that they were opening Grok Bot to a small set of companies this weekend. Slack lit up in under a minute. People asked if we could drop our template if it loaded skills and MCPs. I gave it another thirty minutes on my phone.

Great interface. About a tenth of the power of the setup I already have.

That is the look-a-pony move. A kid at a parade. Every new float is the whole world for ten seconds. Then the next one comes and the last one is gone.

Good engineers have been doing this with AI for two years. A friend tries everything that ships. Another friend burns a Thursday on a bot that cannot see 1Password. I have done it too. The product we were supposed to ship is still sitting there.

You have to try things. If you never try, you wake up on last year’s stack and the people you trust have already moved. If you always try, trying becomes the job.

The Feed Is Not a Curriculum

The story going around is that serious people install every launch. Skip a weekend and you are behind. The person with twenty half-configured connectors looks like they are doing the work of staying sharp.

Two years ago skipping Cursor or Claude cost you. The wow showed up in an afternoon. You could feel the desk change before dinner. The volume is different now. Every week brings another always-on teammate, another model, another demo that can drive your browser. Official Gmail and calendar connectors work. Point the same bot at a custom MCP on your machine and it falls apart. Credentials in 1Password are a coin flip. The demo is real. The afternoon is gone. You call it research because research makes a wasted Thursday feel like craft.

Agents made every interrupt feel free is the cousin about pasting every Slack. This is the other cheap start: the launch tweet looks free, and the bill is the hours plus the work that sat there while you configured someone else’s connector. Code got cheap is judgment once typing is cheap. This is judgment before you open the tool.

What I Actually Do

I do not bite until someone I really trust is getting a wow on real work, not a launch-week wow. Then I give it one to four hours. If I cannot feel the magic in that window, it goes back in the dumpster for a while. That is the whole filter.

Wow is not a polished UI or a friend who clicked around for ten minutes and sent a screenshot. Wow is the desk changing. The thing I already do gets faster, or a thing I could not do becomes possible, and I can feel it without writing myself a pep talk.

Grok Bot is the latest sample, not the point. Official Gmail, calendar, Linear, and Slack work. Custom tools are a coin flip. I pointed it at GitHub and it read my agent. It can see skills. It cannot really inspect them. On mobile, today, it is a thinner copy of the agent I already run. I have tried it twice. Both times it got better. Neither time it replaced the desk. Better later is not a reason to live in it now.

Codex, the first time I tried it, was CLI only and the browser story was bad. An hour was enough to say no. Automations took me about fifteen tries before one version was worth keeping. I come back when the product changed, not when the timeline said I was behind.

The bets that paid over the last two years: Cursor, Claude, Wisprflow, Omarchy, OpenRouter, BMAD. Most did not win on the first try. Some won, lost to the next thing, then came back. I am not married to a logo. I am married to the filter.

At my desk I feel like I can do anything. On a phone I feel hobbled. That gap keeps selling mobile-first bots. A nicer glass rectangle is not a wow if it cannot see the data my work lives in. I am also a bad customer for a lot of this. My stack has been ahead of the default for a while. Pivoting is expensive when the new thing has to beat a setup that already works. If a two-hour try cannot beat your desk, keep the desk.

If your stack is behind, the filter does not flip. You will get more trusted wows because more things will actually change your week. You still write the sentence. You still time-box. You still dumpster the ones that only look current. Being behind is not a license to graze.

What Both Paths Share

Sampling everything and sampling nothing both get called seriousness. One person looks current and leaves a calendar full of half-configured connectors. Another looks stable and runs a museum stack. Both can tell a clean story at standup. Both hide the same miss: they never ran a short, honest try against a written bar.

You still owe a try when a trusted person reports a real wow. You still owe a dumpster when the wow does not show up in a few hours. A contractor on someone else’s toolchain still owes the time box on the tools they control. A junior on a ticket mill still owes the same filter. Missing seniority does not make the feed free. The install is still yours.

How You Practice

Name three people you trust enough to trigger a try. People who do the work, not the launch account. If none of them have used it for real work, you wait.

Before you install, write one sentence for what wow would mean. It has to see my repo and my calendar without me babysitting credentials. It has to cut the thing I already do from four hours to forty minutes. If you cannot write the sentence, you are going in to play.

Give it one to four hours, then decide. A weekend spike with no dumpster rule is how a Thursday disappears. If the magic is not there, write one line about why and put it away. “Interface is pretty, custom tools do not work” is a decision. “I should give it one more night” is how you wake up on Monday with the same empty branch.

Keep two lists: paid-off and dumpster. Date both. Retry a dumpster item when the product changed or a trusted person reports a new wow. Do not retry because you feel behind.

Do not let mobile pain pick the stack. Feeling hobbled on a phone is real. It is a bad reason to throw away a desk that already works. Solve the mobile gap on purpose, or live with it until a trusted wow actually closes it.

Final Thoughts

You have to try things. You cannot let trying dominate your attention.

The feed will keep throwing ponies. A few of them will be worth a stall. Most of them will look free and cost the afternoon. Wait for a trusted wow. Give it a few hours. Keep the ones that change how you work. Put the rest back.

Ready to spend your attention on the work that actually compounds? Join us at MasteryLab.co.