Architectural Designer | Design Systems + AI
Exploring how small studios can use AI without losing their soul.
Experiments, workflows & honest lessons.
This is Part 7, the last one, of a series on building a knowledge system for running a studio. Here's what to actually type once yours exists.
1/ "What did we agree with the structural engineer on the beam sizing, and when?" Tests whether the decision trail holds up before you need it for a real dispute.
2/ "Summarise everything outstanding across all live projects that needs my decision this week." A full triage, instead of five separate email threads.
3/ "A new assistant is joining on Project X. Draft their onboarding brief." Turns the project file into a handover document automatically.
4/ "The client's asking why joinery costs increased since the cost plan. Pull the full trail." Direct answer, instead of digging through months of email.
5/ "I've got a call with the SE in ten minutes. Brief me on the last three exchanges." Meeting prep from the actual record, not memory.
6/ "Draft a fee proposal for this enquiry, based on our last three comparable projects." Your own history as the reference point, not a blank template.
7/ That's the whole series.
Studio Vault Setup Series, Part 7 of 7.
Full article: https://t.co/Def5CakqIr
This is Part 6 of a series on building a knowledge system for running a studio. Here's the step that turns it from a nice filing system into something that actually saves you time.
1/ Claude connects to real tools (Gmail, Notion, and similar) through what's called an MCP connector. Think of it as a plugin that grants read access, and, if you allow it, the ability to act on your behalf.
2/ Set one rule before connecting anything: Claude never sends or deletes anything, whether an email, a message, or a database record, without showing you the exact content first and getting explicit confirmation.
3/ Put that in CLAUDE.md from day one. It's what makes broad read access safe, since nothing happens without your sign-off. [Time: 20-40 min per tool, mostly the login/authorisation step]
4/ Studio Vault Setup Series, Part 6 of 7.
Full article: https://t.co/Def5CajSST
This is Part 5 of a series on building a knowledge system for running a studio. A vault full of good structure is still useless if nobody updates it, here's how that stays automatic.
1/ The fix: a small set of repeatable routines, each triggered by typing a short command.
2/ /context, run first every session. Reads the master files so Claude knows where every project stands before you ask it anything.
3/ /sync, run per project. Searches your email and meeting notes for anything new since the last sync, and appends a dated summary to the project file. This is the one that stops the vault going stale.
4/ /debrief, run once a day or at session end. Sweeps across everything /sync might have missed.
5/ /coach, run whenever you want to think something through rather than update a record. Asks you questions one at a time. Doesn't offer advice unless asked.
6/ Four commands. That's the whole maintenance system. [Time: 30-45 min to set up all four as command files]
7/ Studio Vault Setup Series, Part 5 of 7.
Full article: https://t.co/Def5CakqIr
This is Part 4 of a series on building a knowledge system for running a studio. Your folders exist now, this is what actually goes inside the important one.
1/ Each live project gets its own file. This is where the actual value sits.
2/ Unlike a static overview, this file grows. It ends with a running history section, newest entry at the bottom, nothing ever deleted.
3/ Ask what happened on a project eight weeks ago and you get the real record, not a guess pieced together from memory.
4/ Structure: an Overview section (client, scope, team, status) at the top, then dated updates stacking below it forever. Every session appends, never overwrites. [Time: 15-20 min per project]
5/ This is the part that actually defends you in a dispute. Client asks why a cost moved, or a contractor claims they were never told something: the file has the date, the decision, and the reason behind it.
6/ Not a reconstruction. A record.
7/ Studio Vault Setup Series, Part 4 of 7.
Full article: https://t.co/Def5CajSST
This is Part 3 of a series on building a knowledge system for running a studio. If you've decided you want one, this is where you actually begin.
1/ Everything sits inside one root folder, your vault. Subfolders separate different kinds of information: projects, people you work with repeatedly, clients, daily notes, decisions.
2/ Start with three: projects/, people/, daily/. Add the rest once you actually need them, not before.
3/ The one file that matters most is CLAUDE.md, at the root. It's read automatically every session, before anything else. Think of it as the permanent brief you'd give a new assistant on day one, except you never repeat it.
4/ It should cover who you are, where key files live, rules that never get broken (never send anything on my behalf without showing me first), and how you want Claude to behave.
5/ To start, open an empty folder in Claude Code and paste something like: "I run an architecture studio and want to set up a knowledge vault using Markdown files. Here's my studio and current projects. Create a CLAUDE.md, a folder structure, and one file per live project." [Time: 30-60 min]
6/ Full prompt, ready to paste as-is, is in the article.
7/ Studio Vault Setup Series, Part 3 of 7.
β Part 2: https://t.co/sgBZMeBQYJ
Full article: https://t.co/Def5CakqIr
This is Part 2 of a series on building a knowledge system for running a studio. Start here if you're wondering what problem this even solves.
1/ Run a studio long enough and you're carrying a second job: remembering things. Which detail the engineer changed and why. What you told a client three months ago. Whether the second fix items got confirmed.
2/ Most of that lives in your head, or an email thread nobody can find, or a meeting note nobody read since.
3/ Two failure modes follow. Knowledge walks out the door when staff leave, because it was never written down anywhere durable. And when a dispute arises over who decided what, you're reconstructing from memory instead of pointing to a record.
4/ What I built: a folder of plain text files. Obsidian displays it as a linked graph. Claude Code reads and writes to it directly, so it has full context before it answers anything.
5/ No database, no login screen. Text files, plus an AI that keeps them current.
6/ A normal AI chat forgets everything when you close it. This doesn't, because the memory lives in the files, not the conversation. Every session starts by reading them and ends by updating them.
7/ Studio Vault Setup Series, Part 2 of 7.
β Part 1: https://t.co/b5pV70VCWp
Full article: https://t.co/Def5CajSST
I run an architecture studio.
I got tired of forgetting why a decision was made three months ago, chasing an email thread nobody can find, or watching knowledge walk out the door when someone leaves.
So I built a knowledge system with Claude + Obsidian that actually remembers.
Over the next 7 posts I'm breaking down exactly how it works and how to build the same thing yourself.
Full article here:
https://t.co/DS7ewLKXxu
1. doubling Claude Codeβs five-hour rate limits for Pro, Max, Team, and seat-based Enterprise plans.
2. removing the peak hours limit reduction on Claude Code for Pro and Max accounts.
3. raising our API rate limits considerably for Claude Opus models.
Higher usage limits just got announced and if you missed the last time this happened, this your chance to use Claude to build something new.
Really excited!
https://t.co/51a5SgVHGv
Usage limits are up, effective today we're:
1) Doubling Claude Code's 5-hour limits for Pro, Max, Team and seat-based Enterprise plans
2) Removing peak hours limit reduction on Claude Code for Pro and Max plans
3) Substantially raising our API rate limits for Opus models
@sho_noi Great work! What hit me was your last sentence. The feeling of not enough time!
Are you using Claude to SketchUp via MCP or is this something that you built using Claude code?
Everyone fears AI will replace them. But a graphic designer can build a logo design app with AI - not a floor plan generation app. Why? You need to know if the output is correct. Architects donβt just make drawings plans. We know which ones actually work. Thatβs irreplaceable.
If a tool gives me 10 percent of the output for a task within a live architectural studio workflow, I do not see that as failure.
I see it as signal.
That mindset alone has removed a lot of frustration when testing AI inside real architectural workflows.
The biggest mistake architectural studios make with AI is trying to use it all at once.
The better question is not, βCan AI run this entire workflow?β
It is, βWhich small part of this studio workflow can AI help with today?β