LukeParke/unicorn-utterances
1
0
Fork 0
mirror of https://github.com/LukeHagar/unicorn-utterances.git synced 2025-12-09 12:57:45 +00:00
Code Issues Packages Projects Releases Wiki Activity

WIP

Browse Source
This commit is contained in:
Corbin Crutchley
2021-07-21 14:17:21 -07:00
parent e0e5325896
commit c92e41d319
1 changed files with 40 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

45
content/blog/github-copilot-wont-replace-devs/index.md
View File

@@ -10,14 +10,16 @@
}
---
I recently touched on how GitHub Copilot, an AI-powered code generation tool from GitHub and OpenAI, is going to shift the way we’ll need to do interviews. Copilot frankly is astonishing in its abilities to generate complex algorithm implementations from nothing more than a function name.
I recently touched on how [GitHub Copilot](https://copilot.github.com/), an AI-powered code generation tool from GitHub and OpenAI, [is going to shift the way we’ll need to do interviews](). Copilot frankly is astonishing in its abilities to generate complex algorithm implementations from nothing more than a function name. This makes sense given it's training data of GitHub's publicly hosted community code ([a controversial decision](https://twitter.com/eevee/status/1410037309848752128)).
Some have taken these advanced algorithm assessment capabilities as a warning sign that developers will soon be fully automated using tech similar to Copilot, I’m not sold on that idea.
# Architecture
Let’s first remember what the job of an engineer or developer is. While on the surface, yes, developers do type code into their IDE - the real work is done in the developer’s mind. To code something is to consider a problem’s expected outcome, its constraints, edge cases, and to take those into account to decide on an implementation.
While Copilot is highly capable of generating *a* solution, it doesn’t know your engineering constraints. This is where architecture decisions come into play. Sure, you may know that you want a sorting algorithm - but *which* sorting algorithm may be more important than being able to implement it. After all, if you are wanting to implement a complex sort on a large dataset with limited memory, your biggest problems are likely to stem from knowing where to store your data in an [external sort](https://en.wikipedia.org/wiki/External_sorting) as opposed to the specific code syntax you’ll utilize to make that a reality.
That said, not every engineer is at or needs to be at an architectural level. Some of us are most comfortable when we can focus within our IDEs as opposed to meeting rooms where those constraints often come to light. However, there is a skill that every developer will need to develop as they code: Debugging.
# Bugs
@@ -35,11 +37,15 @@ In order to automate refactors, Copilot would not only need to know how things *
# Code Review
When GitHub Copilot first launched, there was a lot of discussion about how good its generated code would be in the end. Can Copilot understand the nuances in useEffect? Does it know that you need a consistent memory reference to avoid triggering change detection?
When GitHub Copilot first launched, there was a lot of discussion about how good its generated code would be in the end. Can Copilot understand the nuances in `useEffect`? Does it know that you need a consistent memory reference to avoid triggering change detection?
Maybe, but you can’t be certain it will get it right every time. However, the same can be said for others: you can’t be certain another person on the team will get it right every time.
This nuance brings another point against the concept of developers being fully automated by Copilot: Code review.
Ideally, you shouldn't be allowing developers to push code directly to production on a regular basis. While there will always be emergency scenarios where this doesn't apply, it's dangerous to ignore the code review stage. This isn't to say that you shouldn't trust your developers, but we're only human after all. If [Google can make a single-character typo to wipe every ChromeOS laptop with a certain update installed](https://www.androidpolice.com/2021/07/20/a-new-chrome-os-91-update-is-breaking-chromebooks-like-a-bull-in-a-china-shop/), it's not impossible your team may make a similar mistake.
During this process of code review, your team may discover bugs
@@ -49,7 +55,6 @@ This nuance brings another point against the concept of developers being fully a
Further, we’ve had the ability to do something similar for some time now in the form of forum questions. In fact, many have pointed out that this process of looking up code based on its expected constraints is similar to one that a developer might experience by searching StackOverflow for code snippets. Funnily, some thought the idea so similar they decided to build an alternative VSCode plugin to Copilot [that simply looks up StackOverflow answers as suggestions](https://github.com/hieunc229/copilot-clone).
@@ -59,7 +64,13 @@ Further, we’ve had the ability to do something similar for some time now in th
# Copilot’s Strengths
# GitHub Copilot’s Strengths
None of this is to say that Copilot as a tool isn’t advantageous
@@ -67,8 +78,32 @@ None of this is to say that Copilot as a tool isn’t advantageous
I find that developer tools do best for me when they become invisible. Copilot is able to do that quite well and gets out of your way when you don't need it
While [streaming on my Twitch](https://twitch.tv/crutchcorn)
https://clips.twitch.tv/TacitFitIcecreamTriHard-KgJCKYYIEPqxe4dQ
# Conclusion
GCP, I love ya, but your suggestion that [humans need not apply](https://www.youtube.com/watch?v=7Pq-S557XQU) isn’t quite here yet. At least not for developers.
After all, GitHub’s tool is called “Copilot”, not “Autopilot”
GCP, I love ya, but your suggestion that [humans need not apply](https://www.youtube.com/watch?v=7Pq-S557XQU) isn’t quite here yet. At least not for developers.