Plan a bigger job as a chain of prompts
Hiring someone, launching a feature, closing the month: real work is a chain of tasks, not one prompt. Describe the job and get the steps in order, the expert prompt for each, the details to give once for all of them, and each answer carried into the next step.
Read how to break a big job into a chain of prompts, with worked examples.
New to this? What it does, in plain words
One prompt does one task. A job is several, and each one needs what the last one made.
Asking an AI to "help me hire a designer" in one go gets a generic list. Asking it for the scorecard, then the phone screen built from that scorecard, then the feedback written against it, gets work you can use. This plans those steps and hands each answer to the next.
What you need
Nothing. No API key and no sign-up. When you press Plan, your description goes to thinkhard.si, is read, and is forgotten. Everything you fill in after that stays in this browser.
How to use it
- Describe the job in a sentence, or list the steps one per line.
- Press Plan the job.
- Fill in the details once. Every prompt that needs them gets them.
- Open step 1 in ChatGPT or Claude, paste its answer back here, and step 2 is ready with that answer inside it.
The job
Add material it can use: notes, a job posting, a resume, data
Why a chain beats one big prompt
A model asked to do five things at once does each of them less well, and you cannot check the middle. Split the job, check each answer, and pass it on: the scorecard you approved shapes the interview, and the findings you checked shape the report.
- You. Describe the jobA sentence, or the steps one per line.
- thinkhard.si engine. Find the stepsThe order the work happens, from 17 playbooks and 669 expert prompts.Instant
- thinkhard.si engine. Link themWhich answer each step takes, and which can run at the same time.
- You, in any AI. Give details onceThe same detail in five prompts is asked for once.
- You, in any AI. Work down the chainEach answer you paste back fills the next step's prompt.
How the steps are found
- Steps you list. One per line, numbered, or joined by "then", "after that" or "and" before a new task. Each is matched to the expert prompt that does it, read in the context of the whole job, so "summarize the interviews" in a research job finds the research prompt, not the one for workplace investigations.
- A job in one sentence. "We're hiring a designer" or "close the month" is matched to a playbook: the steps that kind of work takes, in the order it happens, with what each step hands to the next.
- Unsure matches show the two nearest prompts, so you can switch.
Details given once
A job posting, a role title or a hiring manager's name turns up in several prompts. The planner asks for each detail once and fills it into every prompt that needs it, along with anything it found in your description or material. What you type here stays in this browser.
Questions
Does it run the prompts for me?
No. Each step opens in ChatGPT, Claude or any AI you use, with its details filled in. You read the answer, fix what needs fixing, and paste it back here, so a mistake in step 2 never flows unseen into step 5.
What jobs does it know?
Seventeen kinds of job as playbooks, from hiring and launches to month-end close, research, sales and job applications, and any steps you list yourself, matched against 669 expert prompts.
Is my description stored?
No. When you press Plan the job, the description and any material go to thinkhard.si, are read, and are forgotten. The details you fill in and the answers you paste back never leave this browser. See the privacy page.
Can I use it from code or an AI assistant?
Yes. POST /api/v1/plan returns the steps as JSON, and AI assistants that use the thinkhard.si MCP server get it as the plan_workflow tool, so an assistant can work through a whole job step by step.