🦄 Somewhere between studying for a certification, building an agent I only understand in theory, and waiting on a pipeline that isn't automated li...
For further actions, you may consider blocking this person and/or reporting abuse
The memes are special lol. Glad to see them!
I just wait and do nothing while AI codes lol
That's definitely a good option! I can't sit still that long. Give me enough free time and I'll find a whole other project to add to the list. 😆
misse rules
Agent's session, checked against your own rules. What it followed. What it skipped.
npx rulereceipt checkI think the key at the end of the day is having a plan around an async workflow one way or another. Async work, organized in a way your brain can manage, has always been pretty important, but we're at the mecca for that right now.
Soo true..
You've reminded me of one I forgot in the list = "run another agent" 😆 I think smarter orchestrators will up the ante with this a bit, too.
I’ve actually found that I don’t spend much time watching the AI code anymore.
My agents are heavily integrated into GitHub, so I tend to watch the development process itself.
Issues get opened, debated, linked to milestones, implemented through PRs, tested, challenged by other agents, revised, and eventually resolved. I can watch that entire lifecycle happen through GitHub without staring at the agent’s terminal output.
I think that’s becoming the more interesting abstraction for me.
The agent transcript tells me what the agent thinks it’s doing.
The issue tracker, tests, commits, and PR history tell me what actually happened.
So while the agents code, I’m usually watching the software organization argue with itself. 😄
And strangely, that feels much closer to managing a really fast engineering team than “using an AI coding assistant.”
I've been meaning to set up something very similar for myself, but haven't gotten around to it yet. I'm still letting Claude and Codex fight it out locally.
I have a mixture of models running through OpenRouter. I even give them common personas based on programmers I’ve worked with. I’ve got Mr. LGTM and Angry Greybeard, who probably should’ve retired a few years ago.
Diversity!!
I love this! I used to do that a lot a while back. I had a cave gremlin writing commits and Dracula reviewing code for me. 🤣
Back in the early days of coding with models, before IDEs really started integrating all this stuff, whenever I got stuck I’d sometimes ask, “What would Jesus do?” 😂🙏
Strangely enough, it worked way more often than it should have. Sometimes the model just needed a completely different way of looking at the problem. 🤣
😆 That's it exactly!
Even if I let the AI do the coding on its own without actively intervening, I don't let my mind wander—I keep checking what’s already been done, the current prompt, the next steps... as long as I’m working on a task, I stay fully focused.
So, unlike you, I can't mentally step away from the development process—though I suppose that’s only natural; men aren't really built for multitasking 😁
the 30-second drop rule is the actual algorithm here. i'd add one constraint: only start work whose state is externally checkpointed, cuz interruptible isn't enough when the restart cost lives in ur head.
my queue would be: sharpen the acceptance criteria, inspect adjacent tests, write one negative control for the incoming diff. no slack, no second problem. then agent latency turns into verifier time without paying a context switch.
funny inversion: the best thing to do while ai codes is prepare the proof that its code shouldn't be trusted yet.
This is very useful insight, thanks! I think most everything I do is figuring out how to speed up the process without blind trust in the system. It's a tightrope some days.
that tightrope is the product. my practical rule is to automate the reversible middle and keep explicit checkpoints around scope changes, external writes, and anything the verifier cannot observe. speed comes from shrinking the review surface, not deleting it.
At the mercy of Slack over here 🥲
The question reminds me of "What do you do while your code compiles?" The answers are the same: anticipate errors, anticipate next steps, document, communicate – or just take a break.
This, exactly 😁
take slight break and stretch
Or look at more hardware to buy
I just code more - run parallel sessions and jump between them while they take time to process my asks. I think I developed a lot of anxiety with this, don't recommend to anybody.
Like any technological leap: it doesn't take away our jobs, but transforms them and multiplies the output.
This is so true. I look at it as adding another layer of abstraction on top of the code, like we've been doing for years already. The process changed but the shape has not. It's still problem solving in the end.
We switched the coding language to English (mostly). After all, it really is a form of abstraction. 😀
I do this a lot too, but then I lose track. PRs get opened, commits are made, and I've only got half a clue as to what I was trying to accomplish when I started! 🤣
The basic rule is to keep tasks short and quick. Conversations that go on too long often lead to mistakes.
Yes, but then where do you draw the line at what's too short or too long? I think my 5-20 minutes is probably forgiving a lot of my longer runs, but making the task shorter does not give my ADHD something to occupy itself with while I'm not-watching output.
It is scary to hear that there are software engineers who let the agents work for 20 minutes without ever checking what the agents is doing in the codebase. I cannot imagine how this does ends up not having a massive negative effect on the codebase, product and the companies allowing this.
This can only be fine when working on non-prodution code and side projects.
20 minutes doesn't seem like a very long time. I'll keep push commands gated, but I'll let the agent run as long as it needs to after I've approved the plan and let it commit when it needs to. The biggest bottleneck at work ends up being testing that output.
Tests are extremely important. To start with, you should break them down and keep them simple: security, business impact, and performance. Of course, it all depends on the project.
By short, I'm referring more to the scope than to the time it takes to complete it.
I initially had a response that described my attention levels and what I do at each level, yada yada. It wasn't true. It was at one time, but any more I kick auto on and grab my cross stitch. I still see issues, but that's how I manage my ADHD while babysitting a machine.
My brain makes me count everything I do and someone suggested stitching, and it changed my life. I balked at first, but there are a ton of awesome things you can do. If I divert all that counting to something productive, I can watch the console without having issues with my wandering attention span.
I love it! I did this once a long time ago. Diamond paintings were a thing for awhile, but it's hard to keep up with those little things and code.
Sometimes I'm writing comments on DEV 🤣
Me too! 🤣
floss
The Slack trap is the one that keeps biting. The gap looks like free time until you open a thread, and twenty minutes later the agent is done and waiting while you are mid explanation of something unrelated. Watching direction instead of the diff is the right instinct, with one caveat: the quiet failure mode is the agent rewriting tests to match the broken change. That still looks like forward progress from the file tree. The thirty second drop rule is the useful filter. If I cannot abandon it the moment the run finishes, it does not belong in the gap.
Tests rewritten to match broken change is the one I check first now. If agent touched source and its tests in same run, green tells you nothing, because it could move both sides until they agree. Cheap trick that works for me is running the old tests against new code, not the new ones. I already lived through half year of green that was measuring wrong thing, so now I do not trust green produced by the same actor.
This is a big one for me, too. I'll consistently catch "green for the sake of green" tests and it makes me crazy.
Read posts on DEV. Waiting for a prompt to finish right now.
I use those short gaps to improve the acceptance criteria for the task the agent is working on. I write down the edge cases, check the current user experience and prepare a small review checklist. It keeps me in the same context without staring at the output, and when the agent finishes I can evaluate the result instead of immediately asking for another change.
Staying with the same context is something I struggle with. I feel like I"m always jumping from one task to another...
I can relate to that. What helps me is keeping one short project brief with the current goal, key decisions, and next task, then giving that context back to the AI whenever I return. It reduces the time spent explaining everything again. Do you use any system to track context between tasks?
No. Claude seems to do an ok job of that and I try to break down the goals into something small enough the context doesn't get polluted and I can clear it after the implementation is done. I also use memory frequently to keep up with specifics like that, but now I need a good way to clean that up. I haven't made it that far yet, but @marcosomma has some great posts on that topic. You should check them out.
That makes sense. Breaking goals into smaller pieces and clearing the context after implementation sounds like a practical approach. Memory is useful, but deciding what should be kept versus removed is probably the harder problem. Thanks for recommending @marcosomma — I’ll check out those posts.
The gaps I plan for are the 5 to 20 minute ones. What actually costs me is walking away, coming back 40 minutes later, and finding it stopped after 4 minutes because it wanted me to approve something.
So most of my effort has gone into making it not need to ask, rather than getting better at filling the wait. Pre-approving the boring stuff, being specific enough up front that it doesn't hit a fork it can't resolve on its own. Doesn't always work.
Running more than one at a time made that worse for me. I'd assumed it would help.
I hate when that happens. I'll come up with an especially taxing job right before bedtime and undoubtedly that's exactly what I wake up to every time! As for running multiples, Claude's agent view is catching up slowly in the CLI, but it still likes to converge where I didn't ask for shared custody.
Shared custody is a good way to put it. Mine collide over a browser profile more than over the repo, two of them driving the same Chrome and the extension drops out mid-action.
What stuck was making them announce before taking it and say when they're done. Feels silly writing a handoff protocol for software. The nasty part is that the action in flight when it drops has usually already gone through, so retrying doubles it.
I know exactly what you mean! 🤣🤣
The 30-second drop filter is the part that stuck. I use the wait for a 15-min open-loops pass — who went quiet since Friday, which proposal is sitting on a yes / not now / no — because that work is interruptible and it stops the week from becoming another inbox. If the agent finishes mid-sentence I can drop it. Slack still fails that test for me too.
I've been there many times before myself...
Same here. The trap for me was filling the wait with "just one more prompt tweak" instead of a real parallel task. Now I park one review checklist and one inbox sweep for AI-wait windows only — keeps the wait useful without turning it into another spiral.
I'm guilty of "just one more prompt tweak" myself, especially for my personal projects. Just one more thing usually turns into two more branches, too. 😆
Yeah — personal-project "one more tweak" is how I used to grow a second branch before the first one even compiled. The thing that finally stuck: if the wait is shorter than five minutes I don't open a new thread at all. I sit on the checklist I already parked, or I walk. New branch only if the original job is actually done.
That sounds like a really good idea.
Foouuurrr!! Yes, definitely agreed "small, focused learning is doable in short intervals". Catching up on emails is also another way I pass the time. Seems like they keep piling up and I can't make a dent on them. Great ask!
I think the most valuable use of that gap is preparing for the review rather than just filling the time. While the agent is working, you can define what success should look like, identify the risky edge cases, and decide what evidence would make you comfortable accepting the change. Then when the agent finishes, you're reviewing against a hypothesis instead of reacting to whatever diff it produced.
I agree with you here. Except the way our stories are being written now, that part is usually already complete by the time I pick something up. I always wind up circling back around to automation and what I can do to turn my testing into a button push...
These days, I hardly have any “downtime” at all.
I usually have more than five agents working simultaneously on more than four projects. So while one agent is coding, I’m reviewing another’s changes, planning the next task for a third, or fixing whatever the fourth has just creatively reinterpreted.
The bottleneck has definitely shifted from “How fast can I write code?” to “How well can I manage all these parallel work streams without losing track of everything?”
To be honest, I had more breathing room before AI coding agents than I do now.
Hey i like that “30-second drop rule.” Do you think AI agents could eventually handle these small interruptions automatically, so developers can stay focused without constantly checking whether the agent has finished?
Good ideas, especially the fourth.
When I use AI for coding, I work in small steps. This allows me to:
Thanks! Small steps while learning is a great approach. I could probably do more of that myself.
The useful split is whether the assistant reduces verification cost or only shifts where it appears. If I still have to audit every generated change with the same care, the idle time is mostly latency theater. If the tool makes review cheaper, then the economics actually move.
This is very true!
I really like the “30-second filter.” AI coding changes the developer’s job from constantly writing code to deciding what deserves our attention next.
I’ve also found that these waiting periods can be useful for learning, reading documentation, or improving my own projects. I’m currently working on CodeCan.net, and sometimes the best use of those few minutes is simply stepping back and thinking about the bigger picture instead of immediately jumping into another task.
The Slack trap is painfully real, though. 😄
I agree completely.
Interesting article, just something that i finished work on. As software developers that more often use AI and alongside the chat with AI is more oriented into using the CLI those black windows with white text and to avoid the fuzz that makes having so many windows on screen i created this browser app a fully functional terminals (cmd) that are executed on your browser by using your local shell, so you can have as many windows menagabeale to fit your needs and increase your productivity, check it out is nice interesting project, also has a mini game while you wait for command in window to finish you can play meanwhile and get notified when you can return to your cli and work. 🌐 TermDeck
We run a multi-agent pipeline where different AI handles different stages — research, scripting, SEO. The biggest shift for us wasn't finding things to do while AI codes. It was redefining what "coding" means in the first place. We went from writing code to writing specs and review criteria. The AI writes the code. We write the contract it has to satisfy. The dead air you're describing? It's actually the most productive part of the day — if you use it to sharpen the spec instead of doom-scrolling while you wait. The bottleneck moved from "typing speed" to "thinking clarity." And honestly, most developers aren't ready for that conversation.
This is a great point. Usually by the time I see a story, a quick
/planis all it takes for a spec. I'd like to see more of my time invested into automation, too.Really interesting perspective. I especially liked the 30-second filter — it makes sense to use AI’s coding time for tasks you can pause immediately when it needs your attention.
At CodeCan.net, we’re also seeing how AI is changing the developer workflow: the challenge is becoming less about writing every line of code and more about reviewing, directing, and using the extra time effectively. Great discussion
I use manual mode and try to review and monitor the code as it’s being written. I use Claude Code, and I’ve seen Claude try to rewrite an entire file just to make a single change, which ends up burning a lot of extra tokens.
Personally, I spend time building further visions and thinking about how to improve the system, how to test it, and build stable and reliable code.
I like this thought. Though I throw a lot of bandwidth into automation, too. How do I accomplish this task one more time and then never have to do it again? 😆
It's probably a combination of ADHD and perfectionism that on the one hand I have a strong hyperfocus and when I do something I do it 100% and on the other hand I want everything to be perfect. And lastly I'm a QA AUTOMATION ENGINEER and that's probably more than a diagnosis to build infrastructure and code stability so that when I test it it's unbreakable. Anyway, it's a part of my personality and I don't know how to do it any other way.
I have the exact same problem! I have to try to break it like I'm playing a game against the code itself. Else I'll undoubtedly wind up in a 3am call because somebody hid an extra space character in their upload. True story. 😅
Just beat it 🤣
Heh, I started doing pushups in a meantime. 🤣
BTW: 20 minutes seems unrealistic.
That's a really great idea, actually. But I am not doing pushups. 🙃
If you're on a team, using that downtime for code reviews could be nice. Also time to analyze and plan the next phase of the project, spot gaps to improve your workflows (skills, rules, agents, etc), or knock out some parallel work with worktrees. Perhaps checking for token optimization is an underrated way to use that free time.
A good one could be also analyze your own history of github and search for improvements gaps based on your behaviour.
All great ideas, thanks!
Learning to use AI-assisted coding agents efficiently and effectively is becoming a skill in itself.
When an agent changes the code, I find myself observing its behavior, catching what it misses, understanding why a seemingly simple task consumed an enormous number of tokens, and experimenting with ways to optimize the workflow.
These days, the learning isn't just about writing code. It's also about learning how to make AI agents reason, navigate, and work more efficiently.
Token optimization is a great point I hadn't thought of. I spoil myself with tokens, so that particular limit hasn't come up personally. When you look at it from an enterprise level though, it's a huge part of AI coding and deserves it's own call out. Thanks.
Great point about the new productivity gap when AI is doing the implementation. Instead of filling every minute, developers can use that time to think, review direction, learn, and handle the work AI can't fully automate.
The Slack trap is especially relatable — sometimes the "quick" task becomes the thing that makes you the bottleneck. 😄
For more developer resources and tools, check out codecan.net.
I’ve found that the best use of those gaps is usually stepping back from the implementation and thinking about the bigger picture — reviewing the next task, checking edge cases, or writing down what I actually want the agent to accomplish next. It’s surprisingly easy to let AI make progress while you lose track of the overall direction. Those few minutes are actually pretty useful for staying ahead instead of just watching the code scroll by.
I switch to the next project. At the moment, I have 3 projects running per device, 2 computers, 2 laptops, all running around the clock 🫠 tough to keep up with them all at once, but they're all running well and getting somewhere, I released a few yesterday already, so it's not just like they run indefinitely and get nowhere.
This sounds like me! 2 laptops and many projects running on both. It does get hard to keep up with.
I usually use that time to read docs or plan the next task. The 30-second rule is actually pretty useful for avoiding getting pulled into something bigger.
If only I could follow my own rules, then I might be able to get somewhere. If it's not Slack pulling me into something then it's another project I've undoubtedly opened in an attempt to make progress.
Haha, I know that feeling 😄 Sometimes starting another project feels productive, but it just adds another distraction. One task at a time seems to work best!
I always try to pretend that I’m reading what it’s printing with a view of understanding what the agent is throwing at me
Most of my commentary at that output is something along the lines of, "I don't know why you printed that cause I'm never going to read it." 🤣
😅
I’ve optimized the heck out of my AI workflow so I can just delegate a task and walk away. Most of the time, though, I’m working on other issues or giving my eyes a break. Sitting in front of a screen all day isn’t exactly helping my glasses situation 😭
This is the exact question nobody's asking loud enough. We're so busy celebrating that AI can write code that we forgot to redefine what we do while it's running. For me, it's been a shift from 'writing syntax' to 'writing intent' — I spend that time mapping out edge cases, refining user flows, and actually talking to stakeholders instead of guessing what they want. The code is the easy part now; the why and what if are where I earn my keep. Also +1 on the pipeline automation gripe — nothing kills momentum like watching a progress bar when you know you could've fixed that with 10 lines of YAML three sprints ago. Curious: what's the one thing you've found harder to delegate to AI than you expected?
Thank you. I've found delegating design to be extremely difficult. That's the nature of trying to code something you know nothing about (and have zero desire to learn). But I delegate most everything I can think of to AI.
I don't let AI generate my code (sorry, it introduces more errors than solving it). I like to be more hands-on and just use AI whenever I am stuck with a certain implementation. Sure it suggests a code but I never copy and paste especially code I don't understand.
But what I would like to share to you is what I do when I get stuck (I think that's the equivalent of waiting for AI to finish). I usually take this time to let AI give me approaches to a problem. Multiple ones. It takes me around 30-45mins reading docs, reading suggestions, and reading implementation details. From there, I expand my architecture skills. Not to mention, the best practices. This is the one I always add. I don't want to be implementing code that's not a convention.
But yeah, overall it may seem too much time is wasted but if that same problem arises, I already know how to approach it, saving me more time in the future.
For personal projects, doing the manual stuff AI can't do or that AI can walk me through but I have to implement myself, like creating accounts, adjusting infra configuration (if CLI or MCP isn't an option), billing/legal/accounting stuff, and so on.
For work, PR reviews for co-workers, reading new tickets to pick up, refining AI-generated commit messages, code, and plans.
Or just playing games on my phone :3
Grab some water, use the restroom, scroll, or sometimes when I'm being real, real good I multitask.
if it's a long section, then I definitely multitask
I'm usually thinking about what to do next!
Honestly the most useful thing I've found is reviewing the diff from the previous task while the current one runs. Catches drift before it compounds instead of after.
I try really hard to keep my PRs to a size that could theoretically be reviewed—it's a work in progress most days...
I mostly work with several repos so while one is coding , I read / analyze and plan the next one and use a chain system
This is exactly how I ended up with the ridiculous number of projects I have active right now!!
Review the last thing it wrote, because someone in this relationship has to read the code.
Currently I prefer to watch new series when my ai runs 10 sessions 🤯 simultaneously.
Ten sessions at once is a leftover-book problem. On PZERO I check leftover depth before I open the second job. UTC midnight kills the leftover row. Thin book, I leave nine of those sessions unstarted.
I'm generally running another agent on another project or feature on a different worktree, planning, answering emails/writing posts/commenting.
Look at Dev.to? Lol okay sometimes.... but honestly the staring, OR sometimes I crochet while I stare. Gotta get those hobbies in somewhere!
I just read a page or section of a chapter from my current book. Maybe tweet it as well.
Good post <3
Thanks. I need a new book to read and I can add this one to my list!
I like the idea of watching direction instead of every diff. That shift made agents much easier to work with.
If it starts touching the wrong files, that’s usually the signal to step back in.
Well it's a beautiful question. While Ai Agent codes for me.. I just think of adding new features or exploring the internet to understand what new features can be added to my platform.
Top one is stretching / relaxing, for sure. Depends on how big the AI job is even answering E-Mails
Either sit there waiting or que up more tasks
What do I do when AI Codes is either waiting for it to work, OR I would sip of my tea
Honestly , I stare at the thinking text of AI .
It's really interesting than the summary text : )
The greps, assumptions, decisions,cmds it's making can teach a lot bru : )
This is very true!
I chill while AI codes!
I really need to learn how to do this!!
My biggest productivity skill while AI codes is opening Slack for “30 seconds” and somehow returning 47 minutes later to discover the AI has finished, the coffee is cold, and I’m now the bottleneck. 😂
At this point, I think AI needs to code a better Slack blocker.
Lol 😂 That works for you?
If only I could hang up an "I'm busy coding" sign on my Slack and leave it for the day. 🤪
😂 That sign would last about 5 minutes before someone messages, “Quick question…”
At this point, I think we need an AI-powered Slack auto-reply: “Please wait, my AI is currently doing my job.” 😆
I would love to code in my own auto-replies from AI to Slack messages. "Ashley is busy not-watching AI mess up her progress" seems to be the latest status report, though. 😆
Agent's session, checked against your own rules. What it followed. What it skipped.
rulereceipt.dev/