Synapse Fusion · AI Field Guide
32 short guides to using AI well at work.
From how AI assistants work and how to ask well, to agents, safety rules and measuring value. Plain words, real office examples and one exercise in each. Free to read here, no sign-up needed.
The guides
Find the guide you need today.
Five parts, read in order or dip in. Each guide takes 4 to 8 minutes and ends with a 15-minute exercise.
Showing all 32 guides.
Part 1 · Foundations
How AI works, and how to ask it well.
01How AI assistants work, and where they failAI assistants predict likely text from patterns, so they write fluently but can be confidently wrong; use them for drafts and check the facts.Everyone · 5 min read02Search versus AI: how AI finds patterns, and how to feed itSearch finds what exists; AI builds an answer from patterns. Gather the right angles first, and the AI has more to work with.Everyone · 8 min read03Choosing a model and a thinking modePick the smallest, fastest model that does your task well, switch on a thinking mode only for multi-step work, and test on your own tasks.Everyone · 5 min read04Prompt engineering: asking wellA good prompt states the role, task, reader, format and limits, then improves through a few rounds of feedback.Everyone · 5 min read05Context engineering: giving AI what it needs to knowGood results depend less on clever wording and more on giving the assistant the right background, examples, documents and definition of good.Everyone · 5 min read06Markdown: a simple format AI reads and writes wellMarkdown is plain text with a few symbols for headings and lists. AI reads it easily, so it is a good format for instructions and context.Everyone · 5 min read07Images, voice and files: giving AI more than textAI assistants can read images, files and speech; they describe and summarise well, but misread fine detail, so check extracted facts and numbers.Everyone · 5 min read08Deep research: when AI researches for youDeep research modes search many sources and write a cited report; give a clear brief, check the citations, and keep the decision yourself.Everyone · 5 min read09Checking AI output before you use itBefore using AI output, check facts and numbers against real sources, watch for red flags, and make sure a named person signs off.Everyone · 5 min read10Data safety: what never to paste, and other rulesUse only approved AI tools, never paste sensitive data without permission, start with made-up examples, and keep a person in charge of important decisions.Everyone · 6 min read
Part 2 · Building
Making AI repeatable: skills, buttons, connectors, specs and small apps.
11Skills and commands: save what works, reuse it in one wordWhen a prompt works well, save it as a named instruction set so you and your team can reuse it in one step.Practitioner · 5 min read12AI behind a button: one-click tools for repeat workTurn a weekly task into a simple tool with fixed instructions, so anyone gets a draft in one click, ready for a person to check.Practitioner · 5 min read13Connectors and integrations: letting AI use your tools safelyConnectors let AI read or act in your documents, email and systems; give each one the least access it needs and keep a person checking.Practitioner · 6 min read14Building small apps with AI, without codingDescribe a small tool in plain words, let AI build it, test with made-up data, and hand anything risky or holding real data to IT.Practitioner · 6 min read15Specs: write it down before AI builds itA one-page spec says what you want, the rules and what "done" means, so AI builds the right thing and you can check it.Practitioner · 6 min read16Choosing the right level: use what you have before you buildMany needs are met by a feature you already have or a good template; build bigger only when frequency, scale, risk or data require it.Everyone · 5 min read
Part 3 · Agents
AI that plans, uses tools and acts.
17What an AI agent is (and is not)An AI agent works towards a goal by planning steps, using tools and checking results, while a person sets limits and approves what matters.Everyone · 5 min read18Agentic work: how your job changes when AI does tasksAs AI takes on tasks, your work shifts from doing every step to briefing, checking and deciding, and the human parts matter more.Everyone · 6 min read19AI that acts for you: browser and computer useSome assistants can now click, type and fill in forms for you, in your browser or on your computer, if you give permission.Practitioner · 7 min read20Multi-agent environments: a team of specialist AIsSeveral AI agents, each with one clear job, can handle complex work better than one generalist, if a person or orchestrator coordinates and checks them.Builder · 6 min read21Subagents: handing a piece of work to a helperA main agent can hand one focused task to a helper agent with a fresh, clean context, which returns only a short result for checking.Builder · 5 min read22Agent-to-agent workflows: hand-offs, rules and checkpointsWhen agents pass work along, clear hand-off notes, firm limits and a person's approval before any real action keep work safe and traceable.Builder · 6 min read
Part 4 · Running AI well
Harnesses, memory, rules, testing and value.
23Harnesses: everything around the AI that makes it reliableThe harness is everything around the AI model, such as instructions, skills, tools, checks and history, and it shapes how reliable your results are.Practitioner · 6 min read24Memory and learning loops: AI that gets better with useAI improves with use when people record mistakes and fixes, turn repeated lessons into rules, and regularly check what the assistant remembers.Practitioner · 6 min read25Hooks: automatic rules that run at the right momentA hook is a small rule that runs by itself when something happens in an AI setup, so important checks never depend on memory.Builder · 6 min read26Guardrails and permissions: what AI may and may not doDecide in advance what AI may do freely, what needs a person's approval and what is never allowed, then review those limits regularly.Everyone · 6 min read27Evaluations: testing AI before you trust a changeAn evaluation is a fixed set of example tasks you rerun before and after any change, to see whether the AI improved or got worse.Practitioner · 6 min read28Measuring the value: hours given backMeasure AI's value honestly by timing a task before and after, counting people and frequency, checking quality, and keeping estimates separate from measured results.Everyone · 6 min read29Sharing what works across a teamA team gets far more from AI when good prompts are saved, named, credited and shown to others, with sensitive data kept out.Everyone · 5 min read
Part 5 · For leaders
Setting expectations, running the programme and leading the change.
30The AI J-curve: why results dip before they riseAI usually pays back only after people, processes and data adapt, so results often dip first. Plan for the dip, shorten it, and review it honestly.Leader · 7 min read31Concurrent working: learn, build and use at the same timeInstead of requirements, build, test, then roll out, run short loops where real users learn, build, use and measure AI together.Leader · 7 min read32Leading AI adoption: what leaders do differentlyLeaders make AI adoption work by setting rules, protecting learning time, using AI themselves, choosing problems well and keeping people accountable.Leader · 5 min read
Prefer paper? Download the complete AI Field Guide as one PDF (1.7 MB). Examples and figures in the guides are illustrative.
Leaders' questions about the law, data and value are answered in Learn about AI.
Your move
You’re already paying for AI. Let’s make it pay you back.
Start with a conversation. Tell us what concerns you about AI in your organisation. We will listen, answer with practical solutions and propose a path forward that fits. Or write to jacques@synapse-photonics.com.
Synapse Photonics SAS
535 Route des Lucioles · 06560 Valbonne
SIREN 106 579 493 R.C.S. Grasse
Prices exclude VAT