01
One idea per card
atomic sections
One card holds one idea, so it can be found alone, cited alone, and changed alone.
In the Brain: every section is a claim, its support, and a permanent id.
Source: Matuschak, evergreen notes
// the engineering
How the Brain is built.
We believe every company deserves a good brain, so here is exactly how we build one: the ten engineering standards we build on, who defines each one, and exactly how the Brain applies it. Two readers, one page. The plain words come first, the engineering second, the source last.
// the standards we build on
Ten dimensions, first-party sources. No vendor certifies brand brains. We build ours to the published engineering standards anyway, in writing, so the AI that wakes in your Brain understands your brand, your design system, and your workflows from the first word.
verified 2026-08-27 · reverify by 2026-11-25
02
Findable by its name
headings as handles
A card’s title tells you what it holds before you open it.
In the Brain: headings are the routing and retrieval keys.
Source: Matuschak, titles as APIs
03
One front door
one machine entry point
A person and an AI start at the same first page and read the same rules.
In the Brain: two doors, one contract.
Source: AGENTS.md, now stewarded by the Linux Foundation, 2025
04
Readable by anything
plaintext, greppable, no hidden state
The whole Brain is plain text a person can open and any tool can search. Nothing hides in a database.
In the Brain: a git repo of markdown is the product.
Source: Karpathy, llm-wiki, 2026
05
Only what matters
smallest high-signal context
The AI gets the smallest set of truth that fully answers the job, never the whole archive.
In the Brain: budgets enforced by failing checks.
Sources: Anthropic, context engineering, 2025 · Chroma, context rot, 2025
06
Loads only when needed
progressive disclosure
Deep material stays out of the way until a task calls for it.
In the Brain: resident, routed, on-demand, linted.
07
Knows what to keep in its head
tiered memory, core vs archive
Core identity stays resident. Everything else waits, filed, until asked for.
In the Brain: resident core, routed chapters, journal and git as archive.
Source: Letta, agent memory, 2026
08
Every fact wears a date
facts carry their time
A fact records when it became true, so a stale one gets caught instead of trusted.
In the Brain: dated sources, an append-only journal, re-verify dates on connectors. Where a date earns its weight, it is there; where it would be ceremony, it is not.
09
It forgets on purpose
deliberate pruning
Removal is an operation with rules, because a Brain that only accretes gets dumber.
In the Brain: the retire operation, the drop-fate rule, the anti-bloat law.
Source: Letta, agent memory, 2026; Anthropic Projects guidance
10
Shown, not just told
rules paired with exemplars
A voice rule travels with an approved example of the rule done right.
In the Brain: the voice law is paired with an approved-exemplar library that grows with every piece a client signs off.
Source: style-transfer research, ACL 2025
// why small
Small on purpose. Structured on purpose.
About two-thirds of 500
is how many simultaneous instructions the best frontier models follow in practice. A brand is that hard case: many rules at once, all binding.
IFScale, arXiv 2507.11538, 2025
Accuracy also falls as context grows, even inside the advertised window, and models recall the middle of a long context least reliably. Small keeps every rule live.
The honest boundary: small and structured, never small alone. A merely squeezed corpus performs worse, and where a big context genuinely wins, single-fact lookup, we happily let your AI use it. The Brain wins where models degrade most: many entangled rules applied together.
Sources: Chroma, context rot, 2025 · Lost in the Middle, TACL 2024
// the fair question
“Why is this better than a Claude Project with my brand guide and a prompt that says be my best brand manager?”
Try it. Load your guide into a Project, and day one looks similar. We mean that.
The difference is every day after. A pasted guide cannot refuse an off-brand draft by naming the rule it breaks. It cannot run your methodology as skills. It cannot tell an approved claim from a hopeful one. It records nobody’s changes, and it goes quietly stale the week the brand moves. And “be my best brand manager” is adjectives; the style research on this page says voice comes from exemplars and enforcement, not adjectives.
A Project is a copy of your brand. The Brain is the governed source the copies come from, and you can load it into as many Projects as you like. The longer answer is in the FAQ.
// the cadence
Dated, and re-verified
Best practice has a shelf life, so this page carries a date and a due date. Verified 2026-08-27; it falls due 2026-11-25. We re-run the research quarterly: the page updates or holds, and the date resets. A page past its reverify date is stale, and the Brain’s own lint says so before a reader has to. Sources move monthly in 2026; ninety days is the honest shelf life.
// the next step
Hold us to it
This page answered how. The definition answers what: What is a Brand Brain? And if you want the same standard applied to your brand, the conversation starts with Kip and Kevin: free exploration, no obligation, a reply within a business day. Open the contact form.