At Michi­gan Labs, our Asso­ciate pro­gram offers col­lege stu­dents and ear­ly-career devel­op­ers the chance to work with a tech­ni­cal men­tor on real projects — usu­al­ly dur­ing the summer.

Our men­tor­ship goes beyond teach­ing pro­gram­ming; it’s about learn­ing how to work effec­tive­ly. Our hybrid work struc­ture helps ear­ly-career devel­op­ers build good habits and a strong col­lab­o­ra­tive foun­da­tion. For devel­op­ers, this includes learn­ing how to research prob­lems, com­mu­ni­cate progress, and iden­ti­fy next steps.

Devel­op­ers have relied on tools like Google and Stack Over­flow for answers. But now there are faster, AI-pow­ered options. Using arti­fi­cial intel­li­gence (AI) tools well isn’t some­thing typ­i­cal­ly taught in school; it’s learned through hands-on experience.

This year, I ensured our Asso­ciates had access to these tools and encour­aged them to exper­i­ment with them. While I’ve been using AI tools for sev­er­al years, our Asso­ciates are at an ear­li­er stage, fac­ing dif­fer­ent chal­lenges and dis­cov­er­ing unique benefits.

In the post that fol­lows, Nick and Ash­lyn (two of our 2024 Asso­ciates) reflect on their expe­ri­ence with AI tools in soft­ware devel­op­ment. Togeth­er, we hope to help oth­ers ear­ly in their careers under­stand and embrace these tools from a beginner’s perspective.

Here are the ques­tions that Ash­lyn and Nick have respond­ed to:

• What ways have you found suc­cess with AI tools?

• How have they impact­ed how you research problems?

• How have they been a chal­lenge or hin­drance to your work?

• What advice would you give some­one using them?

• Any over­all thoughts or obser­va­tions you’ve found interesting?

Ashlyn’s respons­es #

With most tech­no­log­i­cal advance­ments there are two extremes: some believe it will be a ​“sal­va­tion” that solves many prob­lems, while oth­ers fear it will lead to dis­as­ter. How­ev­er, I’ve found that the truth, espe­cial­ly with AI, usu­al­ly lies some­where in between.

AI has quick­ly become part of my dai­ly work as a soft­ware devel­op­er — main­ly through con­ver­sa­tions with col­leagues and clients, and by using tools like GitHub Copi­lot and some­times Chat­G­PT. I’ve had great suc­cess with Copi­lot, which dif­fers from Chat­G­PT by pro­vid­ing con­tex­tu­al answers based on active tabs and spe­cif­ic code snip­pets. It’s like a ​“Google for soft­ware devel­op­ers,” but more effi­cient because it elim­i­nates the need to sift through search results — focus­ing on rel­e­vant solu­tions with­in the project’s context.

Copi­lot also offers code imple­men­ta­tions or pseudocode along­side its expla­na­tions. I find it most effec­tive when com­bined with Google — using Copi­lot as a start­ing point and then deep­en­ing my research through Google’s vast doc­u­men­ta­tion and resources like Stack Overflow.

In addi­tion to its chat fea­ture, GitHub Copilot’s auto­com­plete and sug­ges­tion tool is incred­i­bly use­ful. As devel­op­ers, we often need to cre­ate repet­i­tive func­tions or expand on com­plex struc­tures. Copi­lot excels at rec­og­niz­ing pat­terns, allow­ing it to com­plete these tedious tasks in sec­onds with just a tab. Since time is mon­ey in any field, this fea­ture can be invalu­able when used effectively.

The chal­lenge with AI tools (like GitHub Copi­lot) lies in how we use them. As I men­tioned before, AI is nei­ther a sav­ior nor a threat; it’s sim­ply a tool.

Like any tool, its val­ue depends on how it’s used and the skill of the user. Copi­lot isn’t per­fect — it won’t always pro­vide a cor­rect answer, and some­times it won’t have an answer at all. The biggest risks are depen­den­cy and lazi­ness. Depen­den­cy means giv­ing up when Copi­lot doesn’t have a response, instead of turn­ing to oth­er resources like Google, doc­u­men­ta­tion, or col­leagues. Lazi­ness refers to neglect­ing code review. Copilot’s solu­tions might con­tain errors, both sim­ple and com­plex, that require care­ful review to catch.

My best advice for using AI tools is to under­stand the code before using it. Don’t just copy and paste. Research and ful­ly grasp the code until you can explain it. This approach helps you grow as a devel­op­er instead of rely­ing too heav­i­ly on AI.

Nick’s respons­es #

When I first start­ed learn­ing React Native, AI tools like Copi­lot and Chat­G­PT were incred­i­bly help­ful. They’ve helped me learn new things, fix my code, and write sim­pler code.

When I know what I want to do but don’t know how to do it, I ask the AI for sug­ges­tions. If the response isn’t quite right, I ask fol­low-up ques­tions until it fits my needs. The tools also save me time by quick­ly find­ing and fix­ing syn­tax errors, like miss­ing brack­ets or semi­colons. Addi­tion­al­ly, when I know how to code some­thing but that it will take a lot of time, I can ask the AI tool to do it, and it gen­er­ates the code in seconds.

The biggest ben­e­fit has been the time saved. Instead of search­ing the inter­net for solu­tions or debug­ging for hours, I can move through tasks faster and focus on the big­ger pic­ture val­ue of a project.

How­ev­er, AI isn’t per­fect. It might not always under­stand what you’re ask­ing or the full con­text of your project, lead­ing to incor­rect results. While this can be frus­trat­ing, you can usu­al­ly keep refin­ing your request until it gets it right. Or you can still turn to the inter­net — or col­leagues — for help.

It’s impor­tant that you per­son­al­ly under­stand any AI-gen­er­at­ed code before using it in your project. The worst mis­take is copy­ing code with­out under­stand­ing it, only to face issues lat­er and not know how to fix them. This can waste more time than it saves.

Over­all, I think that AI is a great tool for boost­ing pro­duc­tiv­i­ty, but you must under­stand how the AI-gen­er­at­ed code fits into your project. If you don’t, it might actu­al­ly reduce your pro­duc­tiv­i­ty by cre­at­ing more prob­lems down the line.

We hope Ash­lyn and Nick’s insights help you eval­u­ate AI tools for soft­ware devel­op­ment. If you’re inter­est­ed in learn­ing more about our Asso­ciate pro­gram, we’ve includ­ed reflec­tions from 2023 below:

• Lessons learned from our Asso­ciate developer
• How to pre­pare for Michi­gan­Labs’ Asso­ciate Devel­op­er position
• How to bring order to the chaos of job applications