> For the complete documentation index, see [llms.txt](https://experience-digital-transform.gitbook.io/experience-digital-transformation/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://experience-digital-transform.gitbook.io/experience-digital-transformation/ai-in-government/working-with-ai-in-government.md). # Working with AI in Government A practical methodology for public servants to use AI assistants well, from everyday drafting to larger work, with a human check at every step. This course shows a government team how to work with an AI assistant step by step. You start by getting good answers in a single chat, then set the assistant up for your team, and finally run larger pieces of work where one chat does the work, another checks it, and you make the final call. It is written for public servants, not developers, and assumes no programming. ## What you will learn * [**Module 1: one good chat**](/experience-digital-transformation/ai-in-government/working-with-ai-in-government/module-1-chat-get-good-answers-in-one-chat.md)**.** Learn to ask clearly and to check the answer before you use it. By the end, you will have a few prompts of your own and answers you have checked. * [**Module 2: set the assistant up once**](/experience-digital-transformation/ai-in-government/working-with-ai-in-government/module-2-prepare-set-up-what-the-ai-knows-and-can-do.md)**.** Give it background on your work, standing instructions and skills, so every new chat starts from there. By the end, you will have a short context pack and a project brief for your team. * [**Module 3: one piece of work, start to finish**](/experience-digital-transformation/ai-in-government/working-with-ai-in-government/module-3-orchestrate-one-piece-of-work-start-to-finish.md)**.** Take one piece of your own work through the cycle once, by hand: you write the instruction, one chat does the work, a second chat checks it, and you decide. You leave with one instruction, one report, one review and your own ruling, in a folder and a log you keep using. * [**Module 4: many pieces of work, for days**](/experience-digital-transformation/ai-in-government/working-with-ai-in-government/module-4-orchestrate-running-it-for-days.md)**.** Keep the cycle running: a log that shows where things stand, a list of common mistakes to watch for, and a way to hand over to a fresh chat. By the end, you’ll be able to say where any piece of work stands, and judge whether your team is ready to work this way. * [**Module 5: the same method, with a plugin**](/experience-digital-transformation/ai-in-government/working-with-ai-in-government/module-5-orchestrate-with-the-plugin-build-a-toy-prototype.md)**.** A plugin handles the bookkeeping while you build a small practice prototype, a toy learner registry. Every decision is still yours. By the end, you’ll have a working prototype, its full record, and a short note on what it shows. Each module ends with a page on judgement: what you should never share with the assistant, and which decisions stay with you. **Which modules to take.** If you commission or approve AI-assisted work, for example as a director or head of unit, Modules 1 to 3 are enough. They let you run any AI usage tip in this knowledge base and judge work done this way. If you do the work yourself, for example as head of an ICT unit, project lead or analyst, take all five. Modules 1 to 4 need nothing installed; Module 5 needs the plugin. {% hint style="success" %} **How the course works.** On each page, you read the idea and why it works, try it in the assistant, run the page’s AI usage tip on your own work, and then compare your result with the worked example, set in [Progressa](/experience-digital-transformation/gea/start-here/progressa-the-demonstration-country.md), a fictional country used throughout this knowledge base. Each module ends with a short self-check. {% endhint %} ## All pages
Module 1 · Chat: get good answers in one chat * [1.1](/experience-digital-transformation/ai-in-government/working-with-ai-in-government/module-1-chat-get-good-answers-in-one-chat/1.1-a-drafting-partner-not-an-oracle.md) A drafting partner, not an oracle * [1.2](/experience-digital-transformation/ai-in-government/working-with-ai-in-government/module-1-chat-get-good-answers-in-one-chat/1.2-the-four-part-prompt.md) The four-part prompt * [1.3](/experience-digital-transformation/ai-in-government/working-with-ai-in-government/module-1-chat-get-good-answers-in-one-chat/1.3-read-the-output-before-you-use-it.md) Read the output before you use it * [1.4](/experience-digital-transformation/ai-in-government/working-with-ai-in-government/module-1-chat-get-good-answers-in-one-chat/1.4-judgement-what-stays-out-of-the-chat-and-what-stays-yours.md) Judgement: what stays out of the chat, and what stays yours * [Self-check](/experience-digital-transformation/ai-in-government/working-with-ai-in-government/module-1-chat-get-good-answers-in-one-chat/self-check-module-1.md)
Module 2 · Prepare: set up what the AI knows and can do * [2.1](/experience-digital-transformation/ai-in-government/working-with-ai-in-government/module-2-prepare-set-up-what-the-ai-knows-and-can-do/2.1-build-the-context-once.md) Build the context once * [2.2](/experience-digital-transformation/ai-in-government/working-with-ai-in-government/module-2-prepare-set-up-what-the-ai-knows-and-can-do/2.2-projects-and-standing-instructions.md) Projects and standing instructions * [2.3](/experience-digital-transformation/ai-in-government/working-with-ai-in-government/module-2-prepare-set-up-what-the-ai-knows-and-can-do/2.3-skills-and-connectors.md) Skills and connectors * [2.4](/experience-digital-transformation/ai-in-government/working-with-ai-in-government/module-2-prepare-set-up-what-the-ai-knows-and-can-do/2.4-judgement-what-never-goes-into-the-setup.md) Judgement: what never goes into the setup * [Self-check](/experience-digital-transformation/ai-in-government/working-with-ai-in-government/module-2-prepare-set-up-what-the-ai-knows-and-can-do/self-check-module-2.md)
Module 3 · Orchestrate: one piece of work, start to finish * [3.1](/experience-digital-transformation/ai-in-government/working-with-ai-in-government/module-3-orchestrate-one-piece-of-work-start-to-finish/3.1-two-failures-and-the-rule-that-answers-them.md) Two failures, and the rule that answers them * [3.2](/experience-digital-transformation/ai-in-government/working-with-ai-in-government/module-3-orchestrate-one-piece-of-work-start-to-finish/3.2-write-the-instruction.md) Write the instruction * [3.3](/experience-digital-transformation/ai-in-government/working-with-ai-in-government/module-3-orchestrate-one-piece-of-work-start-to-finish/3.3-the-report-and-the-review.md) The report and the review * [3.4](/experience-digital-transformation/ai-in-government/working-with-ai-in-government/module-3-orchestrate-one-piece-of-work-start-to-finish/3.4-judgement-you-decide.md) Judgement: you decide * [Self-check](/experience-digital-transformation/ai-in-government/working-with-ai-in-government/module-3-orchestrate-one-piece-of-work-start-to-finish/self-check-module-3.md)
Module 4 · Orchestrate: running it for days * [4.1](/experience-digital-transformation/ai-in-government/working-with-ai-in-government/module-4-orchestrate-running-it-for-days/4.1-the-log-status-is-produced-never-remembered.md) The log: status is produced, never remembered * [4.2](/experience-digital-transformation/ai-in-government/working-with-ai-in-government/module-4-orchestrate-running-it-for-days/4.2-the-faults-you-learn-once.md) The faults you learn once * [4.3](/experience-digital-transformation/ai-in-government/working-with-ai-in-government/module-4-orchestrate-running-it-for-days/4.3-handing-over-to-a-fresh-chat.md) Handing over to a fresh chat * [4.4](/experience-digital-transformation/ai-in-government/working-with-ai-in-government/module-4-orchestrate-running-it-for-days/4.4-judgement-every-day-and-what-this-does-not-claim.md) Judgement: every day, and what this does not claim * [Self-check](/experience-digital-transformation/ai-in-government/working-with-ai-in-government/module-4-orchestrate-running-it-for-days/self-check-module-4.md)
Module 5 · Orchestrate with the plugin: build a toy prototype * [5.1](/experience-digital-transformation/ai-in-government/working-with-ai-in-government/module-5-orchestrate-with-the-plugin-build-a-toy-prototype/5.1-set-up-the-project-and-the-first-round.md) Set up the project, and the first round * [5.2](/experience-digital-transformation/ai-in-government/working-with-ai-in-government/module-5-orchestrate-with-the-plugin-build-a-toy-prototype/5.2-decide-before-you-build.md) Decide before you build * [5.3](/experience-digital-transformation/ai-in-government/working-with-ai-in-government/module-5-orchestrate-with-the-plugin-build-a-toy-prototype/5.3-build-one-goal-per-round.md) Build one goal per round * [5.4](/experience-digital-transformation/ai-in-government/working-with-ai-in-government/module-5-orchestrate-with-the-plugin-build-a-toy-prototype/5.4-check-at-the-real-thing-and-write-the-note.md) Check at the real thing, and write the note * [5.5](/experience-digital-transformation/ai-in-government/working-with-ai-in-government/module-5-orchestrate-with-the-plugin-build-a-toy-prototype/5.5-judgement-what-a-toy-proves.md) Judgement: what a toy proves * [Self-check](/experience-digital-transformation/ai-in-government/working-with-ai-in-government/module-5-orchestrate-with-the-plugin-build-a-toy-prototype/self-check-module-5.md)
## How to use this knowledge base Each page has two parts: * **The concept** explains why something works, in plain words, with its source. It applies to any AI assistant, including the one your ministry has approved. * **In the tool** shows what to click, type or set. The examples use Claude, but other assistants such as ChatGPT, Gemini and Microsoft 365 Copilot have the same settings under different names. Because each setting is described by what it does, you can find it in any of them, and their help pages show where. Each page also includes: * a worked example set in [Progressa](/experience-digital-transformation/gea/start-here/progressa-the-demonstration-country.md), the fictional country used throughout this knowledge base; * an **AI usage tip** you can run on your own work. Each tip tells you what to bring, what you’ll get, how long it takes, and what to check before you use the result; * the sources, with links. ## Prerequisites You do not need any technical background: no programming, no command line and no version control. Module 5 asks you to install the plugin and what it runs on; [With the plugin](/experience-digital-transformation/ai-in-government/working-with-ai-in-government/the-starter-folder/with-the-plugin-your-first-project.md) lists them. **What you do need:** | | | | ---------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | **A piece of your own work** | A briefing, a draft or a note you are about to write. The AI usage tips are designed to be run on real work. If you have nothing suitable at hand, follow the [Progressa](https://experience-digital-transform.gitbook.io/experience-digital-transformation/government-enterprise-architecture-gea/developing-a-gov-enterprise-architecture-gea/start-here/progressa-the-demonstration-country) example instead. | | **An AI assistant** | The one your ministry has approved, or any general-purpose assistant. The *In the tool* parts use Claude. | | **Some time** | Each page is a short read plus five to twenty minutes for the tip. A module takes about an afternoon. | **What you need to install:** | | | | ------------------ | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | **Modules 1 to 4** | nothing. As an option, the [ea-plays kit](/experience-digital-transformation/gea/start-here/the-ea-plays-kit.md) packages some tips as skills, which are ready-made instructions the assistant loads when a task needs them. Page [2.3](/experience-digital-transformation/ai-in-government/working-with-ai-in-government/module-2-prepare-set-up-what-the-ai-knows-and-can-do/2.3-skills-and-connectors.md) shows how to install one. | | **Module 5** | the orchestrator plugin and Desktop Commander. [With the plugin](/experience-digital-transformation/ai-in-government/working-with-ai-in-government/the-starter-folder/with-the-plugin-your-first-project.md) explains how to set them up. |

💬

Module 1 — Chat: get good answers in one chatFor anyone who uses an assistant: why AI sometimes makes things up, how to write a good prompt, how to check the answer, and what never to share.Module 1 — Chat: get good answers in one chat

🧰

Module 2 — Prepare: set up what the AI knows and can doSet the assistant up once, so every chat starts well: background on your work, standing instructions, skills and connectors, and what to leave out.Module 2 — Prepare: set up what the AI knows and can do

🔀

Module 3 — Orchestrate: one piece of work, start to finishTake one real task from start to finish: you write the instruction, one chat does the work, another checks it, and you decide.Module 3 — Orchestrate: one piece of work, start to finish

🗓️

Module 4 — Orchestrate: running it for daysKeep the work running over several days: a log of where things stand, a list of common mistakes, handing over to a fresh chat, and a daily routine.Module 4 — Orchestrate: running it for days

🔌

Module 5 — Orchestrate with the plugin: build a toy prototypeUse the plugin to build a small practice prototype, from setting up the project to writing a short note for a senior reader.Module 5 — Orchestrate with the plugin: build a toy prototype

🗂️

The starter folderReady-to-use forms, prompts, a list of common mistakes and one finished example, to copy and reuse.The starter folder

🚩

The demonstration countryProgressa, the fictional country used in every worked example in this knowledge base.https://experience-digital-transform.gitbook.io/experience-digital-transformation/government-enterprise-architecture-gea/developing-a-gov-enterprise-architecture-gea/start-here/progressa-the-demonstration-country
## Where this sits This methodology is the starting point for everything else in this knowledge base. Whichever [knowledge product](https://experience-digital-transform.gitbook.io/experience-digital-transformation/knowledge-product-gallery) you use, with or without an AI assistant, the same habits apply: ask clearly, check the answer, keep what must stay out of the assistant, and make the final decisions yourself. The step-by-step courses (GEA, GIF, the DPI Roadmap and the Building Block Approach) end every page with an AI usage tip, and this is where you learn to run them well. We recommend taking at least Modules 1 to 3 before starting any of them. All the courses use [Progressa](https://experience-digital-transform.gitbook.io/experience-digital-transformation/government-enterprise-architecture-gea/developing-a-gov-enterprise-architecture-gea/start-here/progressa-the-demonstration-country) for their worked examples and follow the same [ground rules](https://experience-digital-transform.gitbook.io/experience-digital-transformation/government-enterprise-architecture-gea/developing-a-gov-enterprise-architecture-gea/start-here/working-with-ai-the-ground-rules) for working with AI. {% hint style="info" %} **Use this knowledge base from your AI assistant.** Every page is also published as plain Markdown, which an assistant can read. (For whoever sets up your assistant: the knowledge base exposes an `llms.txt` file and an MCP endpoint at `/~gitbook/mcp`.) Point Claude, ChatGPT or another assistant at the knowledge base and ask it to *run the AI usage tip of 1.2 with the following context*. The assistant then uses the knowledge base as its reference, as you do. {% endhint %} --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation by asking a question. Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters: ``` GET https://experience-digital-transform.gitbook.io/experience-digital-transformation/ai-in-government/working-with-ai-in-government.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.