Study hall
Glossary / FAQs
Plain-English words and questions for people who do not already live in AI. Same wording as the papers. This page grows when a new paper adds a word or a question.
Explore: Software and Harness.
From: The job picks the model.
Prices: What the models cost.
* This is from Bench's own experience. It is not what any of the frontier models say about themselves. It will change as new models roll out, or as the ones already here get better.
Glossary
- Orchestra
- A few models taking turns, each doing the part they're best at. One writes the plan. One builds. One checks. Not three chat windows with no jobs.
- Frontier model
- One of the current top models from the companies that make them: Anthropic, OpenAI, or xAI.
- Opus
- The Anthropic model for a normal Tuesday. Nearly as good as Fable, costs less, handles most of the work.
- Fable
- Anthropic's top model. Expensive. Save it for building a pipeline or a big multi-day project. Don't use it to knock out one small job.
- Pipeline
- A system that keeps doing the work. Building a video machine for every job is a pipeline. Asking for one video right now is not.
- One-shot
- One quick job. "Check this email." "Make me one video." Then you're done. Not a project that has to keep running.
- Sol
- OpenAI / Codex's top model. Named for the sun. Not an agent's soul — that's the rules the agent lives by.
- Tokens / budget
- The monthly thinking allowance. Like minutes on a phone plan. Bigger jobs use it up faster.
- Adversarial
- A second model looks for mistakes before you trust the work. Don't call it done because the first one said so.
- Chatbot
- A chat window you visit. You ask. It answers. It forgets. It does not run your business overnight.
- Agent
- Software that lives on your Mac, holds your rules, and does work when you're not looking. You visit a chatbot. An agent stays.
- Harness
- The setup on your Mac that lets agents run the work. You own the Mac. We set up the harness. Your people keep working in Bench from the browser.
- Software
- The Bench app in the browser. Boards, files, people, money. It is not the Mac the agents run on.
- Workspace
- The Software plan. Your people log in. The agents still run on the harness.
- Node
- A $99 monthly network or storage device. Nodes do not run a harness or agents and have no harness setup fee.
- Gateway
- Another computer running a harness and agents. Each additional Gateway is $299 a month plus $525 setup.
- Your AI Plan
- An optional $100 monthly plan for eligible accounts that bought Aurelius before provider connections were included. New harness connections are included. You pay provider subscriptions separately.
- Fuel
- Prepaid thinking you buy from Bench when you don't plug in your own lab plan. The tank. Not the Mac.
- Implementation
- The onboarding you buy with the plan. Hours with a specialist to get one real deal closed in Bench on your data.
- Kestrel
- Measurement credits. A credit is a dollar. Not Fuel. You hand the agent a measurement; you get a priced estimate back.
- AccuLynx
- A roofing CRM. Bench runs next to it. We do not replace the name, and we spell it AccuLynx.
- Roofr
- A roofing CRM. Bench runs next to it. Spell the product Roofr.
- Install crew
- The trade people who physically do the installation. On a roof, in a house, or on a site.
- Event
- The one noun Bench stores for the thing on the board. You may call it a job, a claim, or a deal. The label is yours. The key is Event.
- Act
- A hand-off to a different team. Bench has four, always: acquire, sell, fulfill, close. Your words go on those four. There is no fifth.
- Path
- The kind of job a file is. Retail and insurance are paths. A path adds steps to the same board. It never gets a hand-off of its own.
- Track
- Work that runs alongside the file and can stall it. Supplement, denial, permits, financing. Not a board column of its own.
- Surface
- The few steps a person has to touch: judgment and relationships. Inspection. Present and close. Color confirm. Independent QC.
- Shadow
- The steps agents run underneath. Uploads, day-of carrier calls, build plumbing. You don't live there.
- Bench Page
- A finished page at a link anyone can open without logging in. Forward the link and they see the same page you do.
FAQs
Which model should I start with?
When Bench builds, Grok first. When Bench starts a new client on their Mac, one chat. Opus is the model for a normal Tuesday.
What is an orchestra?
A sequence of models, each with one job. Not three chat windows. At Bench, Grok starts, Claude plans, Codex builds, and a second model checks.
When should I use Claude, Codex, and Grok together?
When the job needs a photo, a video, and a page. You do not need three companies to write an email.
What is the difference between Claude and Codex?
Claude explores and designs. Codex builds, checks, and talks to customers. Do not put Claude in front of a customer. Do not ask Codex to figure out what to build.
When do I use Fable vs Opus?
Opus is the model for a normal Tuesday. Fable is for building a pipeline — a system that keeps doing the work, like a video machine for every job. Don't use Fable to knock out one quick job. That's a one-shot. Use Opus for that.
What is a one-shot?
One quick job, then you're done. Check an email. Make one video. Not a system that has to keep running.
What is a pipeline?
A system that keeps doing the work. Not "make me one video right now." A machine that can make the next ones too.
What is Sol?
Sol is OpenAI and Codex's top model — the sun. It is not an agent's soul. Soul is the rules the agent lives by.
Is ChatGPT an agent?
No. ChatGPT is a chat window you visit. It answers and forgets. An agent lives on your Mac, holds the rules you taught it, and keeps working when you close the window.
What is a harness?
The setup on a Mac you own that lets agents run the work. You buy the Mac. We set up the harness. Your people still work in the browser.
Do I buy Software or a computer?
Both. Software is the app your people log into. The harness is the setup on a Mac you buy and own so the agents have somewhere to run.
Is the first harness a monthly charge?
No. The first harness is included with the first agent. Setup is $525 per harness. Each additional computer running a harness is a Gateway at $299 a month. Nodes are $99 monthly network or storage devices without a harness or agents.
Does every person need a Mac?
No. Most people keep the computer or phone they already have and work in the browser. Only the one Mac the agents live on has to stay on.
Do I have to replace AccuLynx?
No. The agents run next to it. If AccuLynx is where your jobs live, it stays there until you decide otherwise.
Does setting up the Mac count as an Implementation session?
No. Setup is a separate one-time $525 charge. That hour doesn't come out of Single, Guided, or Intensive.
Do unused Implementation hours expire?
No. They carry forward.
Do I have to call it an Event?
No. You keep saying job, claim, or deal, and that's what the board shows. Event is the name of the record underneath so one product can serve more than one trade.
Can I add a fifth hand-off for insurance?
No. Insurance adds steps inside sell, fulfill, and close. It doesn't get its own column.
Is a supplement a stage?
No. A supplement runs next to the job and can hold it up, like a permit or financing. Bench calls those tracks. It shows on the file without adding a column.
What do surface and shadow mean?
Surface is the few steps a person has to touch. Shadow is everything the agents run underneath. You'll only see those words on the board.
What is a Bench Page?
A finished page at a link anyone can open without logging in. Forward the link and they see what you see.
Do you call the agents your install crew?
No. Agents are agents. Your office staff are your people. The install crew is the people doing the install.
What Mac do I buy first?
Size the always-on desktop to the business: up to 5 people gets a Mac mini M6 with 32GB RAM and a 1TB SSD; 6 to 15 gets a Mac mini M5 Pro, 18-core, with 48GB and 1TB; 16 to 30 people or two or more offices gets a Mac Studio M5 Max with 64GB and 2TB; 31 or more people on one central Mac gets a Mac Studio M5 Max with 128GB and 4TB. The 18-core chip is about two hundred dollars more than the 15-core. Get the 48GB first. An M6 at 16GB or 24GB, and an M5 Pro at 24GB, will not hold up as the always-on Mac. An iMac with 32GB and 1TB is another option for the first tier if you want the screen built in. A new MacBook Air that runs its own agents is 32GB. 24GB only if you already own that laptop. MacBooks can join later as gateways, but they are never the always-on main Mac.
Can my laptop be the always-on machine?
No. Laptops sleep, and when the Mac sleeps the agents stop. A MacBook can join later as a second machine that's allowed to sleep.
Is a Kestrel credit the same as Fuel?
No. A credit is a dollar for measurements. Fuel is thinking. They're separate bills.
What is Your AI Plan?
An optional $100 monthly plan for eligible accounts that bought Aurelius before provider connections were included. New harness connections are included. You pay provider subscriptions separately.
What is Fuel?
Prepaid thinking you buy from Bench when you don't plug in your own plan. The agents spend it. What they don't spend stays in your tank.
Is the Scaling Brief a free trial?
No. It's a free diagnostic of how your company runs. Bench doesn't offer a free trial of the app.
What gets locked?
The rates on everything you buy and on your expansion, for two years when you pay the Platform year up front.
What is a skill?
A written page for one kind of work: who owns it, the steps, and what not to do. It's what an agent reads before doing that job.