From Python Core to Personal Apps: Working with AI Agents

AI agents can write code quickly, but they can also repeat bad patterns and produce contributions that maintainers never asked for.

Although Python core developer Mariatta Wijaya uses agents in her own work, she sees the challenges that their use creates for open source projects. Welcome to Agents and Engineers. I’m your host, Dan Gerlanc. Today’s guest is Mariatta Wajaya.

Mariatta is an open source maintainer and contributor with over 20 years of software engineering experience. As a Python core developer, she focuses on proving the core workflow and documentation. She’s also an advisor and administrator for Global PyLadies, a co-founder of PyCascades, and a former chair of PyCon, PyCascades, and PyLadiesCon. In this episode,

We discuss how AI-assisted contributions affect CPython maintainers and what new contributors learn by participating in open source, how Mariatta uses plans, documentation, and agent skills to guide coding agents and correct recurring mistakes, and why she builds personal apps with agents, ranging from a scheduling tool to a workout app. Enjoy the show.

Yeah, it’s been it’s been a long there’s been changes in how we we see up like contributions, like AI assisted contributions and honestly as as co as the maintainers ourselves we use AI. We use gen AI and I know like maybe not all of us there are

not just myself, there are other core developers who just use AI to write code, like they don’t co write code by hand anymore. It’s just prompting and uh you know, use agentic workflows. But we also see we also receive contributions from people and mostly what we see is that the contributions from people weren’t

Up to standard. Like they they they don’t review, they don’t adjust AI output before creating PRs. And that is that’s frustrating to us. Like if they if the user doesn’t first sanitize or review or make sure it sounds like human, it’s

I say this like we we really rather engage with real human than

with AI. And I I know this contradicts what I said earlier where we actually use AI um to contribute. The difference is that as the experienced maintainer we know what the projects needs. We have additional expertise knowing like what is actually good good commit messages or good PR body. Um

good code or good practices in how to adjust unit tests and everything like that. What we see the problem is that first-time contributors or new contributors, they just wanted to have something in CPython. And they just made random things that nobody asked, nobody discussed. And that is the part that is

Draining us. Like it’s not that we say, no, don’t use AI at all. But bring yourself to it. And I think people are missing the point of contributing to open source. For me, like the whole process of contributing open source to open source, like learning how to communicate with people.

in the open, learning to ask questions, clarify, you know, the issues, the requirements, learning how to actually write code or how to use Git, those are part, those are the real point of contributing to open source, getting the experience. And when you delegate all of that to agents without actually learning

I I I yeah, that’s really missing the point of it. And so I right now I see chatters even within core team members. Like there are people who said I don’t want to see this kind of AI slop at all, I don’t use it. There are people like me who actually use it. Um and so we’re divided.

Yeah,

what is the solution to a world where people are increasingly going to be using AI for software engineering? Like how can you

triage or make this feasible if there’s just more PRs and such a range in quality, I would think.

Yeah, that’s that’s what I I I don’t know yet. Especially for me, like I don’t write code by hand now and I see myself as more of the if I could call it like architect or designer or like maybe even product owner just coming up with a set of requirements and designs and then the AI is the one writing the code.

And I and I feel our role as human or as developer will evolve. Like I think for the longest time a person in as in a software engineer they start by their f initial first job is coding. And maybe that’s not their first job anymore. They will start by being the product owner or

people who come up with r requirements. a at least that’s how I’ve I’ve seen myself. Like I d no longer write code, but I can come up with technical requirements because I understand code. Because I’ve I understand what a good software stack l should look like. When I I when I see the code

coming up by AIs. There are times that I would see it. It’s just copy pasting existing code. It would replicate even bad pattern that was accepted before. It will keep doing that. And

I can correct the AI output because uh I’m experienced. I know what it should look like. I can come up with an extra prompt saying don’t do that, do this instead. And I I worry for the new newbies who trusted AI output the first time, they wouldn’t know what’s good.

And how how will they learn?

what’s a good software development patterns are. And I I don’t know if there is a solution right now.

Yeah.

Yeah, I

wonder I imagine it’s significantly more challenging if you have this magic box that will just solve your problems for you to sit down and intentionally develop the skills. I I think there will people will still do it, but I imagine there’ll be

A lot more people who don’t were before that wasn’t really an option.

Mm. Yeah. Yeah, I I again I don’t know what’s the solution. I mean I’ve I’ve delegated things to AI things I don’t know. For example, like

Earlier this year I I had to migrate my soft like my web services from Heroku, which is something I’m super familiar with. I’ve I’ve been using Heroku, I’ve deployed apps on it for years. And when there are news about like Heroku is changing, um I tried to find out a different solution.

and I I had to migrate it off to something else. And I I’m like, this is super infrastructure things that I don’t normally have to deal with. And I don’t have the interest in understanding how to deploy something using Terraform or whatnot. I didn’t want to learn, I just want it done.

And so I just use AI, just tell me the commands. Tell me the CLI commands I need to run to move all of this. And it end up like this whole process of migrating something off Heroku took me like a few hours. Like I think almost the entire evening. And

I think if if I didn’t have AI, I it would not be completed because I tried reading the documentation. I come up I would I tried doing all of it myself and I failed. Like I tried it two months ago and I failed. And so I I do see how it can be helpful. Like with AI

Like w without the AI, I tried reading documentation, came up with error, couldn’t figure out how to resolve this error. Like things just doesn’t work on my machine. Like I would spend more time searching stack overflow. Like no, that’s still not what the error I came up with. But now with AI like in four hours, it’s still a lot of prompting, a lot of trial and error, but I get it done.

So that’s where I feel like you know what I don’t need to know these things. I just want something done. That’s been useful to me. And

I don’t know. I’m just sharing story about it, like how I seen

How it can help? Um

Yeah. Yeah, I know

one day I one time I spent three days troubleshooting AWS permissions issue. And I wonder how quickly agents would have solved it today.

Yeah. Yeah and and I think what helped me with this process when I I like the the whole session that that trial and error with AI agent. It because it records my attempts it records like

first it would say, Okay, do this and I did it and I come up with error because it it records all these attempts. At the end of it, I was able to say, Okay, produce me a documentation or a tutorial so that next time you would answer the right you would answer it correctly instead of the trial and error. And I feel that part is also useful, like being able to come up with artifacts

things that I we did wrong during that pairing session with AI. These are this is also a thing that before, if if I was pair programming with another person, we would focus on actually resolving and then somebody in the end would say, Oh, we need to write a documentation about what just happened.

And guess what? No documentation will be written ever, because we forgot. And I felt by actually typing it out prompting that AI, it can help us record the session. It can come up with documentation that in the future human can learn from it too.

How

will you

That’s that’s how I see

it.

Yeah, usually I would have marked down um like f use for instance whenever I do this kind of session AI designing with s with AI, like I would come with requirements. I wanna build this, this is my specs, I want it to be able to do XYZ and not ABC, like I would start my session with this kind of very specific requirements and ask the AI to

come up with more details like of how it would do it. So before I I ask it to actually build the code, I ask it to come up with the plan first. And I would cr ask it to create a markdown doc and I will go through it. I read through top down. and this doc that

the AI created it would go through many, many, many iterations before I eventually say, Okay, this is good build it. And I definitely have a whole like even within one repo that I work on, I would have multiple, multiple handoff docs and I would tell it to like record this is the decisions we made. if there is something change

recorded so that I still have a history of what I decided and why it was, you know, disproved. This approach was proven to be wrong. That’s why we did it a different way. So yeah, I I try to a do markdown documentation at least

locally myself and whenever I feel like okay this doc is something useful to other people as well. Maybe even if I don’t wanna be too op like imposing like this is the right way to do it. You should all follow it. But at least I will share it. Let people see how I approach things. Then I would just share it as a doc’s documentation site. Um I

These days I just prefer using Markdown. I know in the Python community we still by default we use restructured text for documentation and that’s what the CPython team still use. But whenever I don’t have to use restructured text,

Yeah.

I don’t know if we wanna get into argument about markdown versus restructured text right now, but

yeah. I think a lot of times people don’t need the level of detail you need from restructured text. Like especially yeah, even documentation, technical docs like Claude skills or rules about, you know, how to contribute, those do not need

features from restructured text like plain markdown is good enough. So

So

it sounds like you’ll

Yeah, yeah, it maybe I haven’t been doing it publicly. Some of them I did. I put in my whenever I do like my Claude skills are online now, but there are still personal projects where I just have the docs locally to myself. But it yes, it has.

it has all the iterations like V1, V two, V three all recorded and I would I definitely would tell the AI coding tool to like read the handoff first, read these docs first, make sure you follow my preference, my convention and everything. and I I think that’s another useful thing I learned. Like once

Like at first I just I didn’t have any workflow, I didn’t have any cloud skills. But after a while I learned that it keeps making the same mistake again and again. It keeps wanting to do things that I do not want to do. So that’s when I started creating the agent rules. Um saying like when when you do certain things follow this process. So I guess

The reason I I wrote my Claude skills is because it wasn’t doing what I wanted.

Some of them are very specific. I would see that somehow so this is more like Django code base. it would one of the main reasons I saw it is like you know, in a Django web app

Sometimes we need to assign permissions to different group of people and I’ve seen I’ve seen production code bases where they weren’t even doing it right. Like uh they they would be checking users permission by checking like if a user is in group A, like the the group membership basically, and that’s really not how Django wants you to check permissions, but because

it was part of the code base, of the legacy code base. The agent thinks, oh, this is how you do it, it just keeps copy pasting and writing the code that way. And I spotted it and I said, no, you shouldn’t be using you shouldn’t be using this. You should be using a different function. You should be using like user.has_perm kind of thing. So like

It’s it’s one of those things that you you realize like it’s it’s really just learning from your existing code. It doesn’t really know what’s good. It thought you’ve seen this code pattern before, this is the right way and it do it. And it really needs a human to know that’s not how you’re supposed to be doing things.

There are things like that. Like I I I was seeing it was importing in especially in test cases, somehow it was importing Python imports in line within the f the test functions and I I didn’t like it. I had to instruct to say imports go to the top. And again it’s it’s just a result of it saw bad patterns before and it just

continues doing that. there are also examples where

I I I felt it doesn’t try to come up with better solutions.

I don’t I don’t know how to best describe this.

Can’t can come up with specific examples right now. Yeah.

Like it won’t find or like it won’t find

abstractions. Things like that.

Yeah.

Yeah, something like if if I initially ask it, can I do this? It will continue down that path and and not really thinking like actually if you do something else, it will be much simpler. Like it will keep trying to to satisfy you know the initial requirements. Like it would tweak

it’s I don’t know w it’s world I don’t know how to describe it. Like keep trying to find ways to make sure my initial requirement is d satisfied. But once I go through that and I as a human, like I thought, you know what? No. That’s actually not what I really, really want. I want something else. Like it it would not

switch that idea. Like we still we still need to to come up and say, actually that’s not that’s not it. Um, but yeah, I I can’t think of a specific example right now. I know I know if it happened a few times to me. Yeah.

And is that just

you’ll find it from reviewing the code and asking it to explain it?

Uh it comes different ways, yes, from actually reviewing the end result and also like trying it out. Like I I’ve built a lot of web apps for myself. Um like the initial requirements

Sometimes I don’t really think through the u user experience like you know, like oh I I want a big button here or I want this button to do certain things. It it would be only after like I’ve seen the end result that AI produced. I tried it, I and it just felt odd. And I I had to come in and say no, move this elsewhere.

create an intermediary pop-up or something like that. then and I and I think about this whole experience I feel like okay now I’m becoming more like the QA tester and the product owner and no longer the the developer. Like no longer the person who writes code. And

Personally I I’m okay with that. Like I’m I’m okay with not being the one writing the whole line of code. I’m o be okay with being the one just reviewing and giving a handoff and agreeing or saying no. But I I’ve I’ve I’ve talked to some people and they don’t like that idea. They wanna be the one who actually write the code and coming up with the code and

I I respect that. I think it’s just people have different vision for themselves.

Yeah,

I always think at the end of the day, if it’s our job or something like that. Historically, sure, we were being paid to write in the abstract,

I I totally agree to that and I think one way I I just this is how I justify it to myself. I I I tell myself that I’m not being hired to write code, but I’m being hired for my knowledge, for my expertise and experience. if they just want someone to write code.

Then that’s like the new grad, right? That’s me twenty years ago. That’s my job writing code. But as you like even as an engineer, as you move up in your career from junior to senior to leadership, you will find you don’t write as much code. Like you still do t the technical designs, architecture decisions. You do a lot of meetings with people trying to convince your team members or your product teams to, you know

build this but as you move up you will do more of the leadership thing and not actually writing the code and so I felt yeah like I’m not that junior of the developer anymore. I do not need to be writing code and that’s what I think about this whole thing.

Mm-hmm. Mm-hmm.

Exactly. And I and maybe that’s what I felt why I like using AI nowadays. Now at least I do feel like I’m the boss of it. Yeah.

This you know there is this might be not technical, but recently someone told me that um they’re doing their investments using ChatGPT. Like they would ask ChatGPT Um which stock to buy. Okay, I think that is something I’m not comfortable with doing yet. I would ask

AI s everything but technical things, but I don’t know, that is something a l a little bit too far for me. But I don’t know. I still rely on the real human. Yeah. I I still r depend on Yeah. Yeah.

So I would not I would not do that. I started my career in quant finance, so I uh

yeah, I worked I worked in asset management for a bunch of years, so that I would I would not recommend that.

Yeah

Yeah, I think that when it comes to investing and health. Go talk to a real doctor. I although nowadays I I went to a clinic and they said they are using AI, like the doctor said we’re recording this conversation. The AI will take notes. Okay. Um but I I rather I don’t know, in the end I still

I would like to be able to trust the doctor’s judgment then AI if I need AI I could already I don’t need to go to the clinic, I can go open my browser and so the reason I go to the clinic because I want I want, you know, a real medical advice. So yeah.

Well yeah. Well

also the AI doesn’t have access to all of the information that someone who’s seeing you in person can obtain by observing you or doing

Yeah, there’s something less interesting maybe like recipes, like when I do cooking. Now I was seeing AI could come up with his own recipe, like because it would be able I guess it would be able to just go through hundreds of recipes online and come up with something.

Those are the things I still want at least to hear someone else try and cook this same recipe. I I I don’t w I do not wanna made up recipe by random things. So that that’s another thing that I just no, don’t put AI on it.

Yeah,

I feel like that’s you see commercials where ChatGPT is telling people recipes, but it’s also I don’t know how well that I don’t know how well that worked.

Yes,

you don’t wanna be the first experimenter of that recipe ever.

Exactly. Yeah,

yeah.

Yeah.

Right? That that’s yeah. That’s my approach with cooking. Like I I don’t maybe this is how it’s different with coding. Like when I do cooking I want to know for real. I d I do I’m not in the mood of being creative. I just want something yummy come out of this.

but coding so cooking is just a a hobby. Coding though it is my background, is what I studied for in there was a time when I feel like

I I I I enjoy experimenting with code. I enjoy trying to see what breaks. and

I don’t know where I’m going with this. I I felt it’s still something different, um, d with the approach I I with AI.

Well

it’s hard to

Yeah.

Yeah. I I I think part of it is that I did feel like my identity as developer was lost when I delegate everything to AI and that’s something I had to some similar to what I said earlier in the end, like convince myself that at this time, at this stage

my worth is not in how many lines of code I write, but is is my twenty years of expertise, of having built a lot of things. Um

Mm, mm

hmm. Yeah, I I agree with that. Like but maybe that’s the testament that as as a human we we grow. We uh we shouldn’t be the same person twenty years ago. Um yeah.

What

is something that you’ve built with agents and are proud of that might not be immediately apparent to everyone?

I’ve done I’ve built a few different things and I don’t know how to say this like

I’ve built things custom software, custom apps for myself. Not something that I would probably

find it use not something other people would find useful. But I’m proud of it because I get to build kind of a clone or alternatives to popular SAS products out there that people get people use for free. Free but with a catch. Like for example, I built my own QR code generator.

That I hosted on my own website.

This is the kind of thing of course you can g use it for free out there, but whenever you go to uh SAS products, you will be you will see a lot of ads and then you get to use your QR code for free. And similarly I have my own um kind of a Calendly clone. You know, when you have a Calendly product, you get to connect all your various calendars.

And then it will tell you which time you are available for meeting with people. And this is very important to me as a you know community member. I deal with a lot people in a lot of different communities. They always people are always asking when can we meet? I built my own app.

That I self-host. So whenever people ask when I meet, I don’t fill in their doodle form. Because even though Doodle is free, when you go to Doodle, you have there is this big pop-up, you say accept cookies, whatever, and you see a lot of ads. Like I hate that. So I build my own and I don’t I don’t pay for Calendly.

And I own my data. I’m not selling things. I’m not giving away private personal information to third party unless they need it. I also recently built um a workout app. Um, this is something that my husband and I we did. So trying to stay healthy, there is an Android app basically that needs a subscription.

It’s it’s a super simple app. Like you open it, you say I’m gonna work out for thirty minutes, and then it will come up with different moves of like okay for warm up, walk in place for five seconds or something like do push up for thirty seconds, plank for thirty seconds, things like that. It will you know prompt you like time to do this. And I think

The first time we use this kind of app it would have a lot of ads. You have to watch a thirty seconds ad and wait. And then you get to work out. Or you have to pay subscription, I think when we look at it like oh hundred dollars a year. Not a lot, but still like it feels like

It doesn’t need this ad, it doesn’t need subscription. So I spent an evening, I spent a few hours like telling Claude, build me this app. So now I have something. Like it it doesn’t have fancy graphics. It doesn’t have a real person, you know, animation. It’s just stick figures. Fine. Works for me.

Now I have my own app like

yeah, yeah. Like I’ve I’ve built a lot of these kind of things. Things that I just felt I just for principle, like I don’t wanna pay subscription for it. I don’t wanna give out cookies. I don’t wanna be tracked by people that I’m doing these things. So like alright, build my own workout app.

And I these are the things that I think if I were to do it all by myself, it would take a long time. And that’s when people would say, like, you know, it’s taking too much time, might as well just pay the whatever, nine dollars a month subscription. But now that I build it, took me took me a few hours. S I spent four hours of my time. I spent AI tokens.

But now I I get to use it forever. and I feel proud of these kind of things. And I don’t know, I maybe other people are thinking like how they wanna build products that can be used by a lot of people and I just thinking

I’m a very unique person. I have my own I’m very picky about things. I have my own specific workflow, specific needs. I don’t have the bandwidth or to try to come up with something that everybody would need. Like I don’t I don’t wanna deal I don’t wanna try to convince people that my app is the right thing. So I just build it up for myself. Um

Yeah.

Yeah, and that’s that’s that’s the thing that I just I don’t wanna care. It’s okay if nobody else use it but me, I’ll and I think I would actually encourage people to think about it this way too. Like I felt the the SaaS model is maybe

I don’t know is that still the right model? Like you you use free things like free Gmail or I don’t know if I should like free products but at the cost of something else, like at the cost of giving away personal information or at the cost of your data being used some other way that you don’t get to choose.

And maybe just use AI to build something just for yourself.

For you, yeah.

Mm-hmm. Yeah. But I I re I recognize I might be speaking from a position of somewhat privilege of that I get to use AI, but I I know not everybody.

using AI is not free, right?

Yeah, the hope as the models get better seems like the latest Qwen models or you can even r run a pretty good one on but the 27B model, which before was not r not really the case, so I hope we continue to see that and it

AI coding becomes more accessible even if you’re not using a relatively expensive remote solution.

Yeah, I think a lot unless you’re use have an NVIDIA GPU. Most of the po most people doing it are using Macs, but you need a relatively new Mac with a decent amount of RAM still.

My one Mac from twenty twenty isn’t quite isn’t quite getting it. Yeah.

Okay. Yeah,

then probably not for me as well. I think I also just have one.

Uh I I do find AI help me a lot, especially with community organizing. What I do a lot on that is the content creation, like social media promotion things. Whenever I said you know, typically when you host an event, you wanna promote it. You wanna get people coming, right? And there is it’s not enough to simply

create um event page or you know a one time you know meetup page. You really need to have some kind of social media campaign, get the news out. And I used to do this myself by hand. Like I would come up with like

fun or clever things to encourage people to come to the event. Like I need to promote our speaker, creating their poster and everything. And I realize it just doesn’t scale. It’s not sustainable to be done by one person. And really what I enjoy from community is really being actually helping or mentoring people. Like I I I love

speaker mentorship tell teach helping like supporting speakers to to give their first talk I’m not actually interested in becoming social media manager for a team. Like like having to to plan out

Okay, on this day I’m supposed to send out the the announcement about speaker A, and then the next day I’m supposed to send out announcement about another speaker about CFP, about sponsorship. So I I just I use AI to just come up with the plan, come up with the social media post, suggest me what the poster should look like. And

And I feel comfortable doing this because I know it doesn’t even like when it comes to community events, it really doesn’t need to be me who do this. It doesn’t really need to be Mariatta’s true voice to promote the event. It if if someone if I have a volunteer, if a team member saying, I will run the your social media campaign, I will let them do it. But I don’t.

Like this meetup is just myself basically. And this is something I will happily delegate to AI to come up with the plan, with the content, with everything. And I just execute it. I’m still the one posting it. And I’m I’m okay with this. I don’t care if people think all my posts sounds like AI. uh

Like it it doesn’t even come from my personal account. It comes from like the meetup community account. Like it’s it’s the thing that this needs to be done. I don’t have volunteers doing it. So I’m gonna spend AI tokens to do this. Um and it it still helps my community. I did get a lot of like in the last the last meetup we did, it was pretty successful. Um people do

People see the post and they sign up. So I I just have to choose how I want to contribute to the community and I realize social media writing is not what I want.

to do something else. Yeah.

Yeah. And I I that’s how I it’s helping me like again if someone if I do have a volunteer who will come and say I will own this whole social media posting do it. Like right now I don’t

And most most meetup communities I know locally they don’t have that kind of resource as well. Meetups runs on zero dollars. Most meetups I know don’t have sponsorship or anything. And and I feel it’s unfair or unreasonable to expect meetup organizers to know to have

all the skills of fundraising skills, communication skills, technology skills, mentorship skills. It’s is too much. It really has it it really is supposed to be multiple people and yet again and again all I’ve seen is a meetup is only run by one or two people. And that’s that’s where AI will help them.

Mm.

Yeah, I I think it really can help you and I can s I do see I do hear from people saying that it’s it’s unethical or whatnot. I I don’t disagree with that, but sometimes we just need things done. And

AI helps us do things.

Yeah. And especially when it comes to engaging with other people. I will personally I will definitely check their AI contributions guidelines. Like if I I will be very mindful. Like if if they said don’t use AI then I will follow it. I will handwrite things.

um if they allow it, what are the rules constraints and I I will adjust and I think we we need to be respectful of those and I I just hope people like projects or communities they just come come up with rules and regul like yeah rules and guidelines of what’s acceptable, what’s not acceptable. But I think we we can’t forbid

the use of AI, like AI is here.

Yeah, I think any I mean it depends. There may be some projects where adding limiting AI might work, like they’re very stable and there doesn’t need to be a lot of change. But

Any place that

isn’t exactly this is very stable, like you can’t you can’t break the numerical stability of of of NumPy, right? Things like that. Uh

so outside of that I think it’s going to be hard to uh keep people from using it just because there’s there’s a lot of leverage in doing it and

doesn’t mean that

Yeah.

I I think you still need to use AI responsibly. You still need to review the output. Like the I I for me I probably don’t like don’t review like

the re the re the request one by one. Like I will just let it go like, okay, yes, approve all everything. But I I review before the commit is made. I review before the PR is made. I will tell it like this is the commit message I want. And

You you need to be involved. yeah.

Mm-hmm.

Yeah. And that’s that’s something kinda bothers me a little bit. I’ve been involved in some projects. They don’t have any documentations or whatsoever or they just assume as the human we we all know we’re all smart. We know what’s supposed to be good and when I would be I would have pushed them say

You need documentation to write these things. Document this and they wouldn’t they were not doing anything and then when they start adopting AI to their project and now they were like, okay, we need the agents markdown file now. We need to write these things now. And I I feel annoyed that they do this for the machine. They should have done this long time ago for the human. And that’s

That’s also one of my personal principle now. Like I will write for agents, I will have agents MD file, but I want human to understand it too. Um if if I write an agent there will be a human who can read it. And that’s that’s also one of the reasons I publish my Claude Skills as a static site.

something readable in the browser by human we can click it through it navigated uh so that like you said these are useful things for human to know about and so I I didn’t want to stop with just here’s a set of agents MD files for your agents like I want it to be useful for both human and agent so

Uh recently I I started in addition to my Claude Django skills, I started writing my Git skills as well. Although I don’t think it’s super any different than any other best practices in Git skill, but because I’ve because I started creating a lot of, you know, small personal projects for myself and I found that I need to create

repositories with um with certain standards, right? When you create a GitHub repository you need to enable things like I don’t know, you need to enable I I like to enable like the squash squashing in PRs. things like always require CI, always require

hundred percent test coverage and I I used to do all of this in my head. I knew when I create a repo I will set this up. Now my Clot skill for like git for cit skill for COD has that rules written specifically. Like

When when Mariatta says create a new repo, ask do you want it private or public? What is the license? But other things are automatically set up. It will have CI, it will have unit tests. without me having to prompt it again and again and again. So I it it helped out. I just created something, a new repo also yesterday, and it it worked.

So I I think that’s where people should adjust their u their workflow when they use AI. They need to recognize the pattern, they need to recognize what’s what doesn’t work and understand that they can tweak it. They need they need to eventually come up with their own Claude agent skills or whatever.

I I have for I have code and markdown for work. Like sometimes I would come up with like automation for example I I created something for enabling feature flag and optimizely as part of like production you know production deployment. So I I

So that is a script written in as part of the agent skill. Like I would tell the agent, when I say create feature flag, it will just run that CLI. So it will not try to come up when I when I say create feature flag, it will not try to come up with its own one-off script again and again. I’ve seen that happen. Like it it will try to call request.

Python requests and do curl um to access data. So I I noticed that’s how what it was doing and I I said create an automation script, execute that. So I I do a mix of both.

Yeah. Yeah, one thing I would mention like I would share, for example, in my

speaker management skill basically. I I need so one of the workflow when I run a meetup, I need I need to fetch speaker data from s an app like sessionize or s even spreadsheets. And then I need to populate, you know, like my announcement page with all of this speakers information. And I think the first time I tell the agent to say

create an event page for me. It it has its own custom curl like custom requests parsing the s data from spreadsheet and write up this um event description and I realize I don’t want it spending time coming up with this you know

CSV parsing again and again. And so I I I made it create a script that parse speaker spreadsheet into this format. And so next time I need to do this again, yeah, it it knows this script exists. It calls the script instead of coming up with random things. And I I think that’s that’s an important part that

I know this script needs to be written because I I would have done it myself and because I I pay attention to the output, like to what Claude is doing. And I I could imagine if people didn’t realize this this is what was happening, it they just keep accept, accept, accept. They might not realize that their workflow could have been improved.

Yeah, so basically my my um like the PyLadies Vancouver documentation now includes all of these steps along with the sc the script for parsing things from spreadsheets.