{"slug": "why-are-your-ai-automations-still-running-on-your-laptop", "title": "Why Are Your AI Automations Still Running on Your Laptop?", "summary": "A developer argues that AI automations still rely on humans as 'expensive middleware' to connect steps that APIs and agents could handle directly. The post outlines a vision where agents autonomously handle production errors, from diagnosis to deployment, with humans only approving key decisions. It emphasizes the need for persistent memory or state to carry context between agent steps, a gap the developer is exploring through automations on SchemaBounce.", "body_md": "You wouldn’t run your SaaS off your laptop.\n\nSo why are we doing it with AI?\n\nA lot of the AI automation I see right now still depends on a person sitting in the middle moving information around.\n\nProduction throws an error. Someone checks AWS logs, drops them into Claude, figures out what happened and creates a Jira ticket.\n\nAnother dev picks up the ticket and has an agent make the change.\n\nSomeone reviews it. QA tests it. Maybe DevOps has Codex check the deployment logs over MCP.\n\nInformation comes in. A decision gets made. Something happens.\n\nPretty simple.\n\nExcept we keep using humans to connect each step.\n\nWe are basically expensive middleware\n\nMost of these systems already have APIs.\n\nAWS knows an error happened.\n\nGitHub knows what changed.\n\nJira knows what needs to be done.\n\nCI knows whether the tests passed.\n\nThe agent can read the code.\n\nSo why does someone still need to copy the error out of CloudWatch, explain it to an agent, copy that answer into Jira, then have someone else feed Jira back into another agent?\n\nWe’ve automated the tasks without really automating the workflow.\n\nThat’s the part I think is going to change.\n\nWhere do you actually need a human?\n\nThis is probably the harder question.\n\nAnd every team is going to have a different answer.\n\nSome changes should absolutely require approval. Some companies have compliance requirements. Sometimes the agent is just wrong.\n\nAI output is still pretty meh in plenty of cases.\n\nBut it’s getting harder to argue that a human needs to be involved in every step.\n\nFor a lot of work, I think you really need one or two human decisions.\n\nDid we understand what we are trying to do?\n\nAnd later:\n\nDid we actually do it?\n\nEverything between those two questions is fair game.\n\nEven the second one gets interesting when the intent is specific enough.\n\nIf the requirement says this endpoint needs to return X, these tests need to pass, latency can't increase past Y and this error should stop appearing in the logs... a machine can check most of that.\n\nProbably better than someone clicking around dev for ten minutes.\n\nTake a boring production bug\n\nSay a service starts throwing a new error.\n\nThat error could kick off a job automatically.\n\nPull the surrounding logs.\n\nCheck what deployed recently.\n\nLook through the relevant code.\n\nFigure out the likely cause.\n\nCreate a Jira ticket with the useful context already attached.\n\nThen pick that work up, create the branch, make the change, add the test and open the PR.\n\nDeploy it to dev.\n\nRun the E2E tests.\n\nCheck the logs.\n\nSee if the original problem is still happening.\n\nNobody has needed to do anything yet.\n\nMaybe that's where a developer gets involved.\n\nThey get the issue, the change, the test results and what happened after deployment.\n\nApprove it or don't.\n\nIf you want to be more conservative, put a person at the beginning too. Let them approve the diagnosis before spending tokens and changing code.\n\nIf you want to hit the gas, make merge approval the checkpoint.\n\nThe exact line isn't really the point.\n\nThe interesting part is how much work happens before you need that line at all.\n\nThe part we're missing is memory\n\nWe already have good agents.\n\nClaude Code can work through a codebase.\n\nCodex can work through a codebase.\n\nMCP gives them access to other systems.\n\nThe piece I keep coming back to is what happens when one agent is done.\n\nWhere does everything it learned go?\n\nRight now the answer is often a chat window.\n\nOr a terminal.\n\nOr Jira after a developer manually summarizes it.\n\nThat's not great.\n\nThe error, logs, investigation, decisions, code changes, test results and deployment results should live somewhere the next step can use them.\n\nA database would honestly solve a surprising amount of this.\n\nCall it state, context, memory, whatever.\n\nJust stop making a person carry it around.\n\nThis is a lot of what I've been thinking about while setting up my automations on SchemaBounce. I started out thinking about where agents do work. Increasingly I think the more interesting problem is the infrastructure that lets them keep doing work without needing somebody to babysit every transition.\n\nWhich gets me back to the laptop\n\nYou can build almost all of this locally today.\n\nA couple MCP servers, Claude Code or Codex, some AWS access, GitHub, Jira and a little glue code.\n\nI've done versions of it. A lot of developers have.\n\nBut eventually you end up with a pretty funny architecture:\n\nYour laptop has become production infrastructure.\n\nClose the lid and your automation is gone.\n\nVPN drops? Gone.\n\nToken expires? Gone.\n\nSomeone else wants to use it? Have fun recreating your setup.\n\nWant to know exactly why an agent made a decision three weeks ago? Hopefully you kept the terminal open.\n\nOnce these things are actually doing parts of the SDLC, they probably need the same boring stuff we expect from everything else we run.\n\nHosted somewhere stable.\n\nScoped credentials.\n\nPersistent state.\n\nLogs.\n\nRetries.\n\nTimeouts.\n\nVersioned workflows.\n\nAn audit trail.\n\nObservability.\n\nNot because any of that is exciting.\n\nBecause the goal shouldn't be to build the world's coolest prompt running on a MacBook.\n\nThe interesting version is when an error can happen at 2 AM, the system can investigate it, put together a fix, prove what it tested, and have something useful waiting for a developer in the morning.\n\nNo copy/paste required.", "url": "https://wpnews.pro/news/why-are-your-ai-automations-still-running-on-your-laptop", "canonical_source": "https://dev.to/elliot_hutchins_13af77186/why-are-your-ai-automations-still-running-on-your-laptop-1c5p", "published_at": "2026-08-24 23:46:14+00:00", "updated_at": "2026-08-25 00:13:34.939370+00:00", "lang": "en", "topics": ["ai-agents", "ai-infrastructure", "mlops", "developer-tools"], "entities": ["AWS", "GitHub", "Jira", "Claude", "Codex", "MCP", "SchemaBounce"], "alternates": {"html": "https://wpnews.pro/news/why-are-your-ai-automations-still-running-on-your-laptop", "markdown": "https://wpnews.pro/news/why-are-your-ai-automations-still-running-on-your-laptop.md", "text": "https://wpnews.pro/news/why-are-your-ai-automations-still-running-on-your-laptop.txt", "jsonld": "https://wpnews.pro/news/why-are-your-ai-automations-still-running-on-your-laptop.jsonld"}}