Five Tools That Lost On The Ninth Are Top Tier On The Tenth

“AI tools to run your business. 30 tools, ranked S to D, one tier list.” Six tools per tier, each with a two-word job description. S for best overall, down to D for low value.

It is a good list, and the caption is honest about what it is: “I have tested plenty of tools… A few became part of my daily workflow. Others looked impressive in demos but added very little value in practice.”

It was posted on 10 September. The day before, the same page posted a card headed “AI Hype vs Reality”, crowning a winner in eight categories. We wrote about that one yesterday. Put the two side by side and they disagree about nine tools.

Nine tools, two cards, one day apart

The 9 September card put a tool on top and a tool underneath with a crown. The 10 September card sorts thirty tools into tiers. Here is every tool that appears on both.

The second card explains the first

This is not a catch. The resolution is written in the tier list’s own caption, and it is the most useful sentence either card contains:

“Of course, this ranking is personal. The best AI tool is not always the one with the longest feature list. It is the one that saves you time, fits your workflow, and produces results you can actually use.”

That is correct, and it settles everything. A tier is not a property of a tool. It is a relationship between a tool and one person’s week. Move the person and the tiers move with them — which is exactly what happened overnight, because the two cards were asking different questions.

The 9 September card asked: for this one job, which is better? That is a comparison, and Claude Code can win it.

The 10 September card asked: which tools do I reach for in a normal week? That is a different question, and the answer includes ChatGPT at S-Tier even though it lost a writing comparison — because the job it does in that person’s week is labelled “strategy & workflows”, not writing.

Read that way, there is no contradiction at all in five of the nine rows. Cursor can be the thing you open every day and lose a head-to-head on complex refactors. Perplexity can be S-Tier for fast research and be the wrong tool when the answer has to come from your own files. Both cards are true; only one of them says which question it is answering.

The three absences are a different matter, and simpler. Claude Code, ChatGPT Images and Veo won their categories on day one and are not among thirty tools on day two. The likely explanation is mundane — they are features of products already on the list rather than separate products — but if you were choosing from the tier list alone, three of yesterday’s winners are invisible to you.

Diagram: the two questions a tool card can answer

Which gives you a test for any card like this: before you act on a ranking, find the sentence that says whose week it describes. If there is no such sentence, the ranking is still useful — as a list of candidates, never as a verdict.

All thirty, and what the labels are doing

The D-tier reasons are the most useful thing on the card

Read them again on their own: too much editing · needs review · misses context · inconsistent output · feels generic · limited value.

Four of the six are about work that remains after the output arrives. Not “it cannot do the task” — all six of these tools produce something. The complaint is that what comes back needs editing, needs review, needs the context re-supplied, or cannot be trusted to come out the same way twice.

That is the right way to rate an AI tool, and almost nobody says it out loud. The question is not can it produce a draft — everything can produce a draft. The question is how much of your afternoon the draft costs you after it appears. A tool that gets you ninety per cent of the way and needs a careful read is worth less than a tool that gets you seventy per cent and needs none, because reading carefully is the expensive part.

It also explains the Otter disagreement between the two cards without any contradiction. On 9 September Otter was in a comparison it could not lose on function — it is the only one of that pair that records anything. On 10 September it is rated on rework: “needs review”. Both can be true. A transcript you must proofread is still the only transcript you have.

Four honest caveats

1. Both cards are the same author’s, one day apart, and we are comparing them directly. That is fair because they overlap on nine tools, but it is also slightly unkind: nobody writes two cards expecting them to be cross-examined. The point is not that either is wrong. It is that the two together show what a ranking is.
2. The three missing winners probably have a boring explanation. Claude Code, ChatGPT Images and Veo are capabilities inside larger products, and a list of thirty products may simply not have slots for them. We flag it because a reader choosing from the tier list alone would never see them.
3. We did not test any of the thirty. This is a reading of two cards against each other and against their own captions. The D-tier verdicts in particular are one person’s experience, which is exactly what the caption says they are.
4. Both cards are five weeks old. Posted 9 and 10 September, read in October. For the D-tier especially, “inconsistent output” is the kind of complaint a single release can fix. Tool cards decay quickly in both directions, and we have watched one recommend a product that had already shut down.

The one-line version

Five tools lost a comparison on the ninth and sit in the top two tiers on the tenth; three that won are not on the list at all. Neither card is lying. They answer different questions, and only the second one says so: “Of course, this ranking is personal.”

So treat every tier list as a description of somebody’s week, not of the tools — useful as a shortlist, useless as a verdict. And steal the D-tier’s metric for your own: rate what you use not by what it produces, but by how much of it you have to fix. That number is yours, it does not transfer, and it is the only one that decides whether a tool stays.