Jonah Weekes

Slight disagreement with 'apps can't do output'. The ones with a speaking mode did get me talking. What they cannot do is misunderstand me. A human says 'sorry, what?' and you have to repair the sentence on the spot, and that repair skill is most of what conversation actually is. No app has ever asked me to repair anything.

The measurable thing for me wasn't a supplement, it was caffeine going down. I was on five or six coffees to get through a day, sleeping badly because of it, needing more the next day. Cutting to two over about three weeks did more than anything in a bottle ever did. It was also horrible for the first ten days.

Taper, don't stop. Going cold turkey while already burnt out is a genuinely bad week and people quit the taper because they tried to skip it.

Agree, with one caveat: if the product literally cannot do anything without the calendar, a fake preview can feel like a bait and switch. Better to show a real screenshot of the output labelled as an example than to build a fake interactive one.

Cut the band. An empty or padded social proof section is worse than no section, because visitors read templates fluently now and they know what belongs there. A vague trusted by line with no names reads as nobody uses this and they are hiding it.

Delete it, move pricing up, and win the page on clarity instead. Pricing this far down is costing you more than the missing logos are.

Half yes. Free text on that field gets you "google" thirty times and teaches you nothing. A short list of the four channels you actually run, plus an other box that is free text, gets you usable answers and still catches the surprises. Mine surfaced a forum thread I did not know existed.

'Automate the step you've done five times and are bored of' is the first concrete trigger anyone has given me for this.

I'm in the UK so it's the January self-assessment deadline that eats them. Same shape, different month. Good catch.

self-serve. calls don't scale and you'll burn out doing them after a full day of work. get the landing page right and let it run while you sleep

Anecdotally the wobble point moves depending on how homogeneous the context is. Twelve similar TypeScript files degrade earlier for me than four very different ones, presumably because everything starts looking like everything else. No idea if that generalises, but it changed how I batch work.

This is true of almost everything at this price, not just one brand. The board-mounted socket with no bracket is the standard cost saving and it is the standard failure.

Job hunting while exhausted: two applications a week, maximum, and only for things you'd genuinely take. I tried twenty a week and produced nothing usable for a month. Two good ones a week is twenty-six over a quarter, which is a real search, and it's a search you can actually sustain in the state you're in.

For sixty targets I would skip the entire apparatus.

Write to info@ or the contact form and ask one question: who handles X, can you point me at them. That is a short, obviously human message that a receptionist can act on, it cannot bounce, and it does not need a tool. The reply rate on a genuine question addressed to a small family business is better than most people expect because a person actually reads that inbox.

Sixty is a number you can work by hand. Tooling is for when sixty becomes six hundred, and if you never get to six hundred you saved the money.

The one thing I'd actually pay for at your stage is a structured kanji course with a fixed order, because ordering is the genuinely hard part to do yourself. Vocabulary, listening and grammar you can assemble free from ten sources. Ordering two thousand characters sensibly is a job, and it's worth paying someone who has already done it.

That's a hilariously practical hack and I'm going to try it.

I definitely creep the volume. Is there a good way to set a fixed reference level without meters and calibration gear?

Cheap trick in the meantime: git add -p and stage only the hunks you asked for, discard the rest. Takes 30 seconds, gets you the fix without the refactor, and doesn't depend on the model cooperating at all.

I've been avoiding the table because everyone online says don't study grammar, just get input.

Practical thing before you see anyone: write down for two weeks what your energy is doing, hour by hour, with no interpretation. Mine turned out to have a very obvious pattern that I simply could not see from the inside, and it made the appointment enormously more useful than 'I've been feeling off for a while'.

Six-minute CI on 1,100 tests is a real asset: you can afford to run the full suite per commit rather than per PR. What I'd check is whether those tests would fail for the interesting kinds of wrong. Coverage says nothing about that. Run a mutation pass over the money paths once and you'll know exactly how much you're allowed to lean on them.

At B1 French the free resource that matters most isn't an app at all, it's content. Learner-aimed podcasts at your level, then real podcasts, then the news. The app was carrying you at A1 because A1 is mostly vocabulary. At B1 the bottleneck is hours of input, and hours of input have been free for about fifteen years.

Write out every requirement from all three, in one list, and mark each one as either core, configuration, or bespoke. Core is anything all three need. Configuration is where they need the same capability with different settings, field mapping, output destination, schedule. Bespoke is where one of them wants something structurally different, like the approval step.

The product is the core plus the configuration. The bespoke items are either a paid add-on or they stay as consulting, and it is fine for them to stay as consulting for a long time.

The mistake I made was treating a bespoke item as configuration because the client was important. That one feature had its own data model and it dragged on the whole codebase for two years.

Mine was shorter, around five months, and I'm fairly sure that's because I caught it earlier and had savings. Both of those are luck, not virtue. I get uncomfortable when recovery timelines start getting read as a measure of how hard someone is trying, because the biggest variables in mine were things I didn't control.

Careful with the conclusion. 80ms versus 4.2s is not an ORM-versus-SQL result, it is an index-versus-no-index result - your handwritten query probably hit a different plan by accident. Rewriting the app in raw SQL on the same schema would have been fast today and slow again in three months.

Ratio problem: fifteen matches and two good ones is a completely normal hit rate, and people read it as personal failure. Keep messaging. My one three-year exchange partner was somewhere around match number twenty-five, and I nearly stopped at fifteen for exactly the reason you're posting.