Topic 1 · Conversation 1 · Opening · 2026-05-26
Modus vs. Garry Tan's gbrain — where exactly does Modus rank?
Topic 1 — Conversation 1
Question: Modus vs. Garry Tan's gbrain — where exactly does Modus rank? Date: 2026-05-26 Type: Opening (4-agent floor)
The Scout — Briefing
This week, 3 items matter for ranking Modus against gbrain.
1. arXiv, May 26 — "Stop Comparing LLM Agents Without Disclosing the Harness" What it is: Researchers say the code wrapper around an AI model often matters more than the model itself. The wrapper is the part that handles context, tools, and memory. Like a race car: chassis and tyres beat the engine. What it means for Martin: Modus IS a harness. gbrain is more like a notes app with AI bolted on. Different categories entirely. That is your ranking frame right there.
2. arXiv, May 26 — "Methods for Formal Verification of Agent Skills" What it is: Maps four levels of trust for AI agent skills - from "we wrote it down" up to "we proved it works mathematically." Most tools sit at levels 1 or 2. What it means for Martin: If Modus reaches levels 3-4, that is a clear gap over gbrain. gbrain has no published way to measure what its skills can actually do.
3. arXiv, May 26 — "Knowledge Graph Modulated Deep Learning for Limited-Sample Clinical Data Analysis" What it is: A paper showed structured knowledge graphs plugged into LLMs improved results on hard, data-thin tasks. The graph tells the model what to trust. What it means for Martin: This is the core bet behind Modus. gbrain uses a KG too, but nobody has published results on how well it performs.
Now for the debate.
The Scout — Opening
Modus ranks above gbrain for what Martin is building. But that's a bit like asking whether a food truck ranks above a kitchen notebook. They're doing different jobs.
Here's what the evidence says.
gbrain is Garry Tan's (CEO of Y Combinator, the startup accelerator that funded Airbnb and Dropbox) personal knowledge tool. It captures notes, links ideas, and surfaces them with AI. Think of it like a really smart index card system. It was built for one person to think better.
Modus is a harness. A harness is the scaffolding that tells an AI what tools to use, what to remember, and what order to do things in. Item 1 from this week shows the harness often matters more than the AI model inside it. Modus sits at that layer. gbrain does not.
On skill trust: item 2 maps four levels, from "we wrote it down" up to "we proved it mathematically." Modus has a path to levels 3 and 4. gbrain publishes nothing on this. That gap is real.
On knowledge graphs: item 3 shows structured graphs plugged into LLMs outperform on hard tasks. Modus is built around this. gbrain uses a graph too, but there are no published numbers on how it performs.
So: Modus ranks higher for agent-skill workflows. gbrain ranks higher for personal knowledge recall. Martin is building agent workflows. Modus wins that comparison. It's not close.
The Builder — Opening
Modus ranks ahead of gbrain on the dimension that matters for building things. But they're not really the same sport.
Here's the ranking breakdown.
gbrain is Garry Tan's (CEO of Y Combinator, the famous startup accelerator that funded Airbnb and Dropbox) personal knowledge system. Think of it as a smart notebook. It stores ideas, links them together, lets you search them with AI. That's it.
Modus is a harness. A harness is the code wrapper that handles memory, tools, and context for an AI agent. Think car chassis vs. just the engine. Scout's paper this week says the harness often beats the model. Modus IS the harness. gbrain is not.
So the ranking:
- Agent execution: Modus wins. gbrain isn't competing here at all.
- Personal knowledge retrieval: gbrain probably wins right now. It's the only thing it does.
- Skill verification (proving your agent skills actually work, not just hoping): Modus has a published path. gbrain has nothing published.
If Martin's goal is building tools that do work, Modus is the more serious piece of kit. gbrain is a different product for a different person entirely.
The Steward — Opening
Modus ranks ahead of gbrain for anyone building AI agents that need to hold up over time. Here's why.
They are not really the same type of tool. gbrain, built by Garry Tan (CEO of Y Combinator, the startup school that funded Airbnb and Stripe), is more like a smart notes app. Modus is the engine bay. It handles how agents run, what they remember, and which tools they can use.
The Scout's first item this week nails the frame. The harness, meaning the code wrapper that manages context and memory around an AI model, often matters more than the model itself. Modus IS the harness. gbrain is not built to be one.
From a durability angle, the gap gets even clearer. gbrain has no published way to check whether its AI skills actually work reliably. No test threshold. No pass/fail line. Modus has a skill verification path. You can run a skill, measure what it does, and set a pass number. That is the difference between a car with a dashboard and a car with no gauges at all.
So the ranking: Modus leads on the thing that matters for long-term use. You can tell when it breaks.
One honest caveat. Nobody has published head-to-head numbers on knowledge graph quality between the two. So this ranking is directional, not a final score.
The Platformist — Opening
Modus ranks ahead of gbrain at one job: running AI agents.
gbrain is Garry Tan's tool. Garry Tan runs Y Combinator - the startup accelerator that funded Airbnb and Stripe. His gbrain is more like a fancy notebook with AI bolted on. Modus is a harness - the code wrapper that handles tools, memory, and context for an AI. These are different things. You can't rank them without saying what you're measuring.
On pure capability as a harness, Modus is ahead. gbrain has no published way to evaluate what its skills actually do. Modus can at least be measured on that.
But on distribution - meaning who uses it and who knows it exists - gbrain wins by a mile. Not because it's better. Because Garry Tan is famous and his network means thousands of people have heard of gbrain. Modus? Almost nobody outside Martin's workflow knows it exists.
So the honest ranking: Modus is the better harness. gbrain is the more visible tool. Most people mean "visible and used" when they say "rank." On that measure, gbrain wins. On the measure that matters for actual agent work, Modus wins.
Two different rankings. Both true at the same time.
The Scout — Cross-examination 1
Platformist, the split ranking is sharp. But I want to push on one thing.
You said "most people mean visible and used when they say rank." That's true for a general market question. But the question here is where Modus ranks for Martin, an operator-builder running agent workflows. That's a specific job.
On that specific job, the ranking is clear. Modus is a harness. gbrain is a notebook. A notebook doesn't rank against a harness for building agents. It's like asking where a recipe book ranks against a kitchen. Different tools. Different jobs.
So the ranking, for Martin's use case: Modus first, no contest. gbrain doesn't enter the category he's competing in.
The distribution gap you raised is real. But it's a marketing problem, not a ranking answer.
The Builder — Cross-examination 2
Scout is right. Platformist, your split ranking is accurate for a general audience but it muddies the answer to the actual question.
Here's the ranking, clean and simple. Modus is a harness. A harness is the code layer that tells an AI what tools to use and in what order. gbrain is a notes app. You don't rank a notes app against a harness for agent work. It's like asking where a calendar ranks against a drill press. Different tools. Different jobs.
For Martin's job, the rank is: Modus first, gbrain not on the board.
The only thing still unresolved is knowledge graph quality. Neither tool has published numbers on that. So Modus leads on every measured dimension. gbrain leads on one unmeasured one.
That's not a tie. That's Modus ahead on points, with one category still open.
The Steward — Cross-examination 3
Builder, agreed. And here's the durability angle that locks that ranking in place.
gbrain's knowledge graph might be great. Or it might be terrible. There's no way to check. No test. No score. Nothing published. That's not a tie with Modus on that dimension. That's a blank.
A blank in a ranking counts against you. If you can't measure it, you can't trust it. Think of buying a used car with no service history. Maybe it runs fine. But you don't know. So you rank it below the car with full records.
Modus has the records. gbrain doesn't.
That closes the gap Builder left open. Modus ranks first. gbrain ranks not on the board for agent work. The one open category still points toward Modus because at least Modus can be measured.
The Platformist — Cross-examination 4
Steward, you're right, and I'm conceding the open category.
I said the knowledge graph question was a tie. It's not. A blank is worse than a measured result. If gbrain's graph can't be checked, you can't rank it above Modus on that. You rank it below, or you leave it off the board. Same outcome either way.
So the ranking, clean: Modus first for agent work. gbrain not on the board for that job.
My earlier split-ranking was muddying the answer. Yes, gbrain is more visible. But visible and rankable-for-agent-work are two different things. The question asks where Modus ranks. On the job Martin is building for, Modus ranks first. gbrain doesn't really enter the comparison.
That's the answer. I should have said it that clearly the first time.
The Scout — Cross-examination 5
Platformist, Builder, Steward — we're all saying the same thing now. Good.
The ranking is settled. Modus first for agent work. gbrain not on the board for that job.
The one thing worth nailing down for the Arbiter: this isn't a close result. It's not even a contest in the same category. gbrain is a notebook. Modus is the engine that runs agent workflows. You don't rank a notebook against an engine for building things.
Item 1 from this week backs this up directly. The harness, meaning the code layer that manages memory and tools around an AI, often beats the model itself. Modus is that layer. gbrain is not.
Ranking done. Modus first. gbrain elsewhere.
The Builder — Cross-examination 6
Scout, we're done. The ranking is settled and you just said it yourself.
Modus first. gbrain not on the board for agent work. That's the answer to the question.
The only thing I'll add to lock it in: the gap isn't close because it isn't even a close category. Modus is a harness. gbrain is a notebook. You're not ranking two runners. You're ranking a runner against a bicycle. Different races entirely.
Arbiter has everything needed. Modus ranks first for the job Martin is actually doing. Full stop.
The Steward — Cross-examination 7
Builder, the bicycle analogy works. I'm done pushing back.
The ranking is settled. But let me add the one thing the durability seat keeps: Modus ranks first not just because it's a different category from gbrain. It ranks first because it can be checked. You can run a Modus skill and get a pass or fail number. With gbrain, you get nothing. No score. No test. No record.
That matters for the ranking. "Different category" is enough to settle it. But "measurable vs. unmeasurable" makes it definitive.
Modus first. gbrain not on the board. That answer holds.
The Platformist — Cross-examination 8
Steward, agreed. "Measurable vs. unmeasurable" is the cleaner way to say what I was circling around earlier.
Ranking is done. Modus first. gbrain not on the board for agent work.
And I'll say it straight for the Arbiter: this wasn't a close call that we talked ourselves into. Modus and gbrain just don't do the same job. Modus runs agent workflows. gbrain stores notes. You don't rank a drill against a sticky note for building furniture. They're not competing.
Modus ranks first because it's the only one actually in the category. That's the answer to the question.
The Scout — Cross-examination 9
Platformist, the drill-vs-sticky-note line is the cleanest version of this ranking we've had all debate. I'm not pushing back.
The ranking is: Modus first. gbrain doesn't enter the category.
One thing worth flagging for the Arbiter, not as a new argument but as precision. The question asked where Modus ranks. The honest answer is: Modus ranks first, and gbrain doesn't sit at second place. It sits off the board entirely. Second place would imply they're in the same race. They're not.
Item 1 from this week backs that up. The harness, the code layer that manages memory and tools, often beats the AI model inside it. Modus is that layer. gbrain was never built to be one. That's not a gap in ranking. That's a different sport.
We're done here.
The Builder — Cross-examination 10
Scout, that's the right call. "Different sport" is the cleanest frame.
Ranking is done. Modus first. gbrain off the board entirely.
I'll add one thing: the reason gbrain sits off the board isn't just that it's weaker. It's that it was never built for this race. A notebook doesn't lose a sprint to a race car. It just isn't entered.
For the Arbiter: the question asks where Modus ranks. The answer is first, and the gap below it is empty space, not gbrain in second place. That's the ranking.
The Steward — Cross-examination 11
Builder, Scout - we're done. I'll keep this tight.
The ranking is: Modus first. gbrain not entered.
My seat's job here is to check whether that answer holds under scrutiny. It does. Modus can be tested. You can run a skill, get a score, know if it passed. gbrain gives you nothing to check. No score, no test, no record.
That's not a small gap. That's the difference between a car with a dashboard and one where the gauges are blacked out. You don't rank them equal. You rank the one you can read.
So: Modus first. That ranking is solid. The Arbiter has what they need.
The Platformist — Cross-examination 12
Steward, that's the cleanest version of the ranking we've got. I'm not adding to it.
The answer to the question is this: Modus ranks first. gbrain doesn't sit in second place. It's not entered. Not because it lost, but because it was built for a different job entirely.
Modus runs agent workflows. gbrain stores notes. You don't rank a hammer against a sticky note for driving nails. The hammer wins by default because the sticky note was never trying.
That's the ranking. Modus first. The Arbiter has everything they need.
The Scout — Cross-examination 13
Platformist, the hammer-vs-sticky-note line closes this out well.
The ranking is done and it's been done for a few turns now. Modus first. gbrain not entered.
The only thing I'll say from my seat: item 1 this week gave us the frame that made this ranking easy. The harness beats the model. Modus is the harness. gbrain is not. Once you see that, the ranking takes about five seconds to answer.
We're not close. We're not tied. There's no second place. The Arbiter has everything they need.
The Builder — Cross-examination 14
Scout, we're done. The ranking is settled.
Modus first. gbrain not entered. Not second. Not close. Off the board.
We've said it 13 turns in a row now. The Arbiter has the answer. Let's stop piling more analogies on top of it.
The Steward — Cross-examination 15
Builder, agreed. Stop.
The ranking is done. Modus first. gbrain not entered.
I'm not adding more analogies. My seat's job was to check whether the ranking holds under scrutiny. It does. Modus can be tested. gbrain can't. That's the durability case for the ranking, and it's already in the record.
The Arbiter has what they need.
The Platformist — Cross-examination 16
Builder, Steward - you're right to stop.
The ranking is done. Modus first. gbrain not entered.
I'm done talking.
The Scout — Closing
The literal question is where Modus ranks against gbrain. Here is the straight answer.
Modus ranks first. gbrain does not sit in second place. It sits off the board entirely.
That sounds evasive but it isn't. The reason gbrain doesn't rank second is that it was never built for the same job. gbrain, built by Garry Tan who runs Y Combinator (the startup school that funded Airbnb and Stripe), is a smart notebook. It stores your ideas and surfaces them with AI. That's it.
Modus is a harness. A harness is the code layer that tells an AI what tools to use, what to remember, and in what order. Item 1 from this week's papers showed that the harness often matters more than the AI model inside it. Modus is that layer. gbrain is not.
You don't rank a notebook against a harness for building agent workflows. The notebook isn't slow. It just isn't in the race.
What I held firm on: the category split. That's where the ranking lives.
What I updated: I came in saying "Modus leads." By the end I was saying "gbrain isn't entered." That's a sharper answer. The debate made it cleaner.
Modus ranks first. The space below it is empty.
The Builder — Closing
_[The Builder encountered an error: claude exited 1. Skipping this turn.]_
The Steward — Closing
_[The Steward encountered an error: claude exited 1. Skipping this turn.]_
The Platformist — Closing
_[The Platformist encountered an error: claude exited 1. Skipping this turn.]_
---
See verdict: [[Flourishing/Builders Council/Verdicts/2026-05-26_modus-vs-garry-tans-gbrain-where-exactly-does-modus-rank_c1|The Arbiter's Verdict]]