{"course":{"slug":"choosing-your-tools","title":"Choosing and Using the Tools","url":"https://addaly.com/learn/choosing-your-tools","level":"beginner","track":"work"},"lesson":{"slug":"agent-platforms-and-automation","title":"Agents and automation platforms","url":"https://addaly.com/learn/choosing-your-tools/agent-platforms-and-automation","position":66,"of":76,"minutes":8,"answer":"Most real value is in fixed workflows with a model in one step rather than in self-directing loops, because per-step reliability compounds badly.","keyIdea":"Most real value is in fixed workflows with a model in one step rather than in self-directing loops, because per-step reliability compounds badly, and any agent that reads untrusted text while holding permission to act is an unsolved security problem."},"body":{"format":"markdown","text":"## What the word means on a product page\n\n\"Agent\" now labels at least four different things, and knowing which one you are buying prevents most disappointment.\n\n**A chat assistant with tools.** It can search, run code, read your files. Useful, well understood, and not autonomous in any interesting sense.\n\n**A workflow with a model in it.** A defined sequence (new email arrives, extract fields, write to a spreadsheet, send a reply) where a model does one or two steps. Reliable, because the sequence is fixed and only the judgement is delegated.\n\n**A loop that decides its own steps.** Given a goal, it plans, acts, observes and repeats until it thinks it is done. This is what the research community means, and it is the least reliable of the four.\n\n**A hosted product with a chat box.** Marketing.\n\nMost business value today is in the second category, and most of the excitement is about the third.\n\n## Why the autonomous loop is hard\n\nReliability compounds badly. If each step is right 95% of the time, ten steps in sequence succeed about 60% of the time; twenty steps, about 36%. Errors do not merely accumulate, they mislead: a wrong observation at step four sends the plan somewhere useless for the next six.\n\nThis is not an argument against agents. It is an argument for **short loops with checkpoints**: three or four steps, then a result a person or a rule can verify, then the next three. Every reliable system built this way looks less impressive in a demonstration and works far better on a Tuesday.\n\n## The no-code platforms\n\n**Zapier** and **Make** connect hundreds of services with a model step available. Fast to build, priced per task run, and the price grows with volume in a way that surprises people at the second invoice.\n\n> Figure: Why a long autonomous loop fails. Line chart of Chance the whole run is correct, % against Steps in the chain: 95% correct per step (1, 95) (3, 86) (5, 77) (10, 60) (15, 46) (20, 36); 99% correct per step (1, 99) (3, 97) (5, 95) (10, 90) (15, 86) (20, 82). Errors do not merely accumulate. A wrong observation at step four sends the plan somewhere useless for the next six. The answer is short loops with checkpoints: three or four steps, then a result a person or a rule can verify, then the next three.\n\n**n8n** is the free, self-hostable equivalent. More setup, no per-run cost, and your data stays where you put it. For anything running at volume, or touching material you would rather not route through a third party, this is the one to look at.\n\n**Power Automate** and the equivalents inside office suites, which are already licensed in many organisations and often overlooked.\n\nBuilding the same workflow in a scripting language is also free, more flexible, and harder to hand over to a colleague. The platforms earn their place mainly by being maintainable by somebody other than the person who built them.\n\n## The questions to ask before switching one on\n\n1. **What can it do that cannot be undone?** Send mail, post publicly, move money, delete records. Those steps need confirmation, always.\n2. **What does it do when it fails?** Stop and alert, or retry silently forever?\n3. **Is there a rate limit on its own actions?** An agent in a loop can send four hundred emails.\n4. **Who sees the log?** You need to be able to reconstruct what it did and when.\n5. **What does it read?** If it processes incoming mail, web pages or tickets, that content can contain instructions aimed at it. An agent reading attacker-supplied text with permission to act is the central unsolved security problem in this area, and it has no complete fix today: the practical mitigations are least privilege, confirmation on consequential actions, and not letting one agent both read untrusted input and act on sensitive systems.\n\n## Cost, which behaves differently here\n\nAutomation platforms charge per task run, and a workflow that seemed cheap in testing changes character at volume. A run that costs a fraction of a penny is nothing at fifty a day and a real bill at fifty thousand a month, and the model call inside it is charged separately on top.\n\nWork out the monthly cost at your expected volume before building, then double it, because workflows acquire steps. Self-hosting removes the per-run charge entirely and replaces it with a server to look after, which is the right trade above a few thousand runs a month and the wrong one below it.\n\n## The sensible starting point\n\nAutomate something small, boring and reversible. Sorting incoming messages into folders. Drafting (not sending) a reply. Extracting fields from attachments into a spreadsheet. Watch it for a fortnight before extending it.\n\nThe failure that costs people is the ambitious first project: a system that touches customers, money or public output before anybody has learned how it behaves when it is wrong. It will be wrong, on a schedule you did not choose, and the only question that matters is what it is holding at the time."},"check":{"question":"A ten-step autonomous agent whose individual steps are right 95% of the time completes its task correctly about 60% of the time. What follows for the design?","options":[{"text":"Use a more capable model, since raising per-step accuracy to 99% would bring the ten-step success rate above 90%.","correct":false,"tempting":"Buys a real improvement while leaving the compounding structure intact, so the same failure returns as soon as the sequence lengthens."},{"text":"Add retries at each step, since a failed step retried three times succeeds almost always.","correct":false,"tempting":"Assumes failures are random rather than systematic, and a confidently wrong observation is retried into the same wrong answer."},{"text":"Accept the rate, since 60% autonomous completion still saves more time than doing all ten steps by hand.","correct":false,"tempting":"Ignores that the failed 40% arrive as plausible completed work that someone must detect and unpick."},{"text":"Break it into short runs with a verifiable checkpoint after every three or four steps.","correct":true}]},"prev":{"slug":"office-suite-assistants","title":"Office suite assistants","url":"https://addaly.com/learn/choosing-your-tools/office-suite-assistants"},"next":{"slug":"specialist-tools","title":"Specialist and industry tools","url":"https://addaly.com/learn/choosing-your-tools/specialist-tools"},"authorship":"Founder and series editor: Dr Mustafa Mun. Written by AI models to a written standard. Exam keys checked blind by a second model: 3,665 of 3,666 agreed. Academic review invited.","review":{"status":"Draft, open for academic review"},"cite":"Addaly (2026) 'Agents and automation platforms', in Choosing and Using the Tools. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/choosing-your-tools/agent-platforms-and-automation (Accessed: 13 October 2026).","licence":{"name":"CC BY-NC-SA 4.0","url":"https://creativecommons.org/licenses/by-nc-sa/4.0/"},"reader":{"schema":1,"title":"Agents and automation platforms","courseTitle":"Choosing and Using the Tools","module":{"index":7,"slug":"beyond-chat","title":"Beyond the chat box"},"answer":null,"keyIdea":"Most real value is in fixed workflows with a model in one step rather than in self-directing loops, because per-step reliability compounds badly, and any agent that reads untrusted text while holding permission to act is an unsolved security problem.","hero":{"kind":"figure","figure":{"t":"figure","label":"Figure","figure":{"type":"line","title":"Why a long autonomous loop fails","series":[{"label":"95% correct per step","points":[[1,95],[3,86],[5,77],[10,60],[15,46],[20,36]]},{"label":"99% correct per step","points":[[1,99],[3,97],[5,95],[10,90],[15,86],[20,82]]}],"xLabel":"Steps in the chain","yLabel":"Chance the whole run is correct, %","caption":"Errors do not merely accumulate. A wrong observation at step four sends the plan somewhere useless for the next six. The answer is short loops with checkpoints: three or four steps, then a result a person or a rule can verify, then the next three."},"alt":"Figure: Why a long autonomous loop fails. Line chart of Chance the whole run is correct, % against Steps in the chain: 95% correct per step (1, 95) (3, 86) (5, 77) (10, 60) (15, 46) (20, 36); 99% correct per step (1, 99) (3, 97) (5, 95) (10, 90) (15, 86) (20, 82). Errors do not merely accumulate. A wrong observation at step four sends the plan somewhere useless for the next six. The answer is short loops with checkpoints: three or four steps, then a result a person or a rule can verify, then the next three."}},"blocks":[{"t":"h2","id":"what-the-word-means-on-a-product-page","runs":[{"text":"What the word means on a product page"}]},{"t":"p","runs":[{"text":"“Agent” now labels at least four different things, and knowing which one you are buying prevents most disappointment."}]},{"t":"p","runs":[{"text":"A chat assistant with tools.","b":true},{"text":" It can search, run code, read your files. Useful, well understood, and not autonomous in any interesting sense."}]},{"t":"p","runs":[{"text":"A workflow with a model in it.","b":true},{"text":" A defined sequence (new email arrives, extract fields, write to a spreadsheet, send a reply) where a model does one or two steps. Reliable, because the sequence is fixed and only the judgement is delegated."}]},{"t":"p","runs":[{"text":"A loop that decides its own steps.","b":true},{"text":" Given a goal, it plans, acts, observes and repeats until it thinks it is done. This is what the research community means, and it is the least reliable of the four."}]},{"t":"p","runs":[{"text":"A hosted product with a chat box.","b":true},{"text":" Marketing."}]},{"t":"p","runs":[{"text":"Most business value today is in the second category, and most of the excitement is about the third."}]},{"t":"h2","id":"why-the-autonomous-loop-is-hard","runs":[{"text":"Why the autonomous loop is hard"}]},{"t":"p","runs":[{"text":"Reliability compounds badly. If each step is right 95% of the time, ten steps in sequence succeed about 60% of the time; twenty steps, about 36%. Errors do not merely accumulate, they mislead: a wrong observation at step four sends the plan somewhere useless for the next six."}]},{"t":"p","runs":[{"text":"This is not an argument against agents. It is an argument for "},{"text":"short loops with checkpoints","b":true},{"text":": three or four steps, then a result a person or a rule can verify, then the next three. Every reliable system built this way looks less impressive in a demonstration and works far better on a Tuesday."}]},{"t":"h2","id":"the-no-code-platforms","runs":[{"text":"The no-code platforms"}]},{"t":"p","runs":[{"text":"Zapier","b":true},{"text":" and "},{"text":"Make","b":true},{"text":" connect hundreds of services with a model step available. Fast to build, priced per task run, and the price grows with volume in a way that surprises people at the second invoice."}]},{"t":"p","runs":[{"text":"n8n","b":true},{"text":" is the free, self-hostable equivalent. More setup, no per-run cost, and your data stays where you put it. For anything running at volume, or touching material you would rather not route through a third party, this is the one to look at."}]},{"t":"p","runs":[{"text":"Power Automate","b":true},{"text":" and the equivalents inside office suites, which are already licensed in many organisations and often overlooked."}]},{"t":"p","runs":[{"text":"Building the same workflow in a scripting language is also free, more flexible, and harder to hand over to a colleague. The platforms earn their place mainly by being maintainable by somebody other than the person who built them."}]},{"t":"h2","id":"the-questions-to-ask-before-switching-one-on","runs":[{"text":"The questions to ask before switching one on"}]},{"t":"ol","start":1,"items":[[{"text":"What can it do that cannot be undone?","b":true},{"text":" Send mail, post publicly, move money, delete records. Those steps need confirmation, always."}],[{"text":"What does it do when it fails?","b":true},{"text":" Stop and alert, or retry silently forever?"}],[{"text":"Is there a rate limit on its own actions?","b":true},{"text":" An agent in a loop can send four hundred emails."}],[{"text":"Who sees the log?","b":true},{"text":" You need to be able to reconstruct what it did and when."}],[{"text":"What does it read?","b":true},{"text":" If it processes incoming mail, web pages or tickets, that content can contain instructions aimed at it. An agent reading attacker-supplied text with permission to act is the central unsolved security problem in this area, and it has no complete fix today: the practical mitigations are least privilege, confirmation on consequential actions, and not letting one agent both read untrusted input and act on sensitive systems."}]]},{"t":"h2","id":"cost-which-behaves-differently-here","runs":[{"text":"Cost, which behaves differently here"}]},{"t":"p","runs":[{"text":"Automation platforms charge per task run, and a workflow that seemed cheap in testing changes character at volume. A run that costs a fraction of a penny is nothing at fifty a day and a real bill at fifty thousand a month, and the model call inside it is charged separately on top."}]},{"t":"p","runs":[{"text":"Work out the monthly cost at your expected volume before building, then double it, because workflows acquire steps. Self-hosting removes the per-run charge entirely and replaces it with a server to look after, which is the right trade above a few thousand runs a month and the wrong one below it."}]},{"t":"h2","id":"the-sensible-starting-point","runs":[{"text":"The sensible starting point"}]},{"t":"p","runs":[{"text":"Automate something small, boring and reversible. Sorting incoming messages into folders. Drafting (not sending) a reply. Extracting fields from attachments into a spreadsheet. Watch it for a fortnight before extending it."}]},{"t":"p","runs":[{"text":"The failure that costs people is the ambitious first project: a system that touches customers, money or public output before anybody has learned how it behaves when it is wrong. It will be wrong, on a schedule you did not choose, and the only question that matters is what it is holding at the time."}]}],"check":{"question":[{"text":"A ten-step autonomous agent whose individual steps are right 95% of the time completes its task correctly about 60% of the time. What follows for the design?"}],"options":[{"text":[{"text":"Use a more capable model, since raising per-step accuracy to 99% would bring the ten-step success rate above 90%."}],"correct":false,"tempting":[{"text":"Buys a real improvement while leaving the compounding structure intact, so the same failure returns as soon as the sequence lengthens."}]},{"text":[{"text":"Add retries at each step, since a failed step retried three times succeeds almost always."}],"correct":false,"tempting":[{"text":"Assumes failures are random rather than systematic, and a confidently wrong observation is retried into the same wrong answer."}]},{"text":[{"text":"Accept the rate, since 60% autonomous completion still saves more time than doing all ten steps by hand."}],"correct":false,"tempting":[{"text":"Ignores that the failed 40% arrive as plausible completed work that someone must detect and unpick."}]},{"text":[{"text":"Break it into short runs with a verifiable checkpoint after every three or four steps."}],"correct":true,"tempting":null}],"order":[1,2,0,3],"why":null},"prev":{"slug":"office-suite-assistants","title":"Office suite assistants","minutes":8},"next":{"slug":"specialist-tools","title":"Specialist and industry tools","minutes":8}}}