engineering house, debates about whether or not utilizing LLMs in coding is an effective or a nasty thought are raging. On the extremes, there are some individuals who suppose that any use of LLMs in coding is indefensible, and different persons are able to throw out the entire discipline of software program engineering as too costly and pointless. I’m fascinated by this dialogue, notably from my sociologist’s perspective, as a result of a lot of the dialog appears to be about whether an LLM is useful. Is it so fatally flawed that it’s a complete waste of time, or is it so unimaginably highly effective that it’s a idiot’s errand to keep away from it? As a substitute of enthusiastic about it so narrowly, I really actually wish to speak in regards to the broader context of software program engineering within the context of LLM know-how. (I think about software program engineering solely adjoining to my discipline of machine studying engineering, however I’ve labored alongside many extraordinarily proficient SWEs in my profession, at tech firms of assorted sizes, and I’ve had the prospect to look at the position fairly intently. A few of the evaluation under additionally applies to the MLE discipline, however not all of it.)
Context
Throughout my very own profession I’ve seen quite a bit change within the American tech financial system, and in the way in which the job of “coding” has been perceived and described. It’s a little bit of a trope but it surely was actually true for a time period that “studying to code” was touted because the quickest option to a white collar profession with a gentle paycheck and future-proof prospects, each for younger folks and for career-changers looking for a spot within the shifting financial system. Coding bootcamps came and (mostly) went, for a time giving folks a comparatively fast option to acquire the foundational abilities of the job.
In lots of fields, nonetheless, shortages of certified staff can swiftly flip right into a glut of expertise with nowhere to go (ask attorneys), particularly when excessive or not less than snug salaries are on supply. Whereas some folks suppose that is taking place in software program engineering, it’s not clear if that’s essentially the case but, as a result of there are different components in play too. Within the mid-2010s, many tech firms and startups had been scrambling for hyper-growth and hiring as quick as potential, utilizing low-cost enterprise capital to fund their expansions with out regard for profitability, however that’s definitively over now. Tech layoffs, together with many software program engineers, exploded within the years instantly following the beginning of the COVID pandemic, largely as a response to the unsustainable hiring of previous years and the sudden jolt to the financial system of lockdowns. After that interval of layoffs, hiring for SWE expertise slowed in lots of sectors. Developer skillsets are diversifying and specializing, so there should still be sub-fields with robust hiring, however numerous younger SWEs particularly are struggling to interrupt in.
Junior vs Senior
This brings us round to now. Some folks argue that there aren’t too many software program engineers, however there are too many inexperienced or junior software program engineers, and that employers are nonetheless determined for skilled hires. What does that basically imply? Is it true? With out doing educational analysis to essentially examine (if anybody has executed good analysis on this query, please ship it to me!), we will nonetheless check out the problems concerned. First, let’s outline phrases.
I’m not utilizing anyone’s in-house leveling phrases, or speaking about job titles right here, however I wish to make clear what the variations is perhaps in somebody’s position relying on the expertise and abilities they convey.
A junior SWE is somebody who wants fairly a little bit of assist, teaching, and managing to supply good high quality work. They’re not but at a spot of working independently or mentoring others. They’re nonetheless studying how the job is completed. This doesn’t imply they don’t produce worth! Junior engineers do write code and get work executed, they usually can usually be taught in a short time and turn out to be extra productive. Nonetheless, a senior SWE is somebody who will be anticipated to work principally independently, with minimal supervision required for good work to be executed and objectives to be met, and they are often relied on to assist these junior of us alongside. They begin to have the ability to remedy tougher issues, write elegant code that is top quality, and even start being a part of technique conversations about methods to construct your complete codebase for the lengthy haul.
This is kind of the sample for many expert professions — you’ll be taught some quantity of what you could know from education or coaching, however you actually discover ways to be a member of the career and do the job from expertise and follow. There’s no set rule about how a lot time it takes so that you can go from junior to senior on this framework, as a result of it’s all about what you are able to do and the capabilities you’ve got. Your perceived worth as a employee (and your pay) go up as you get higher on the job over time.
LLMs Enter Stage Left
With this, we will begin to inquire about how that shifts, and the way the existence of LLMs and LLM primarily based coding instruments really might (to make use of a cliche) disrupt the context of the career now.
As a person, in the event you’re a senior SWE, you would possibly discover an LLM software like Github Copilot or Cursor useful for doing scut work, reminiscent of getting unit assessments working, writing fundamental docs, or sketching out the beginning of a challenge that you simply’ll then fill in primarily based by yourself information. Nonetheless, your personal information is actually important for this to be efficient. As we in machine studying will let you know, LLM output is probabilistic at its core. If the coaching knowledge has many examples of code which might be related to what you’re doing, it’s going to be moderately prone to produce code for you that works. However ultimately (in all probability sooner) it’s going to generate one thing that received’t work. That is perhaps failing loudly or failing silently, it is perhaps opening up a safety gap, or it would simply be a suboptimal approach of doing a activity. The factor that issues at that time is what you do in response.
Assuming you’re a senior SWE with important expertise writing all these things by hand your self, you’ve got the abilities to identify the errors, and work out methods to repair them easily so the top product works effectively. The LLM in all probability makes you extra productive ultimately, since you might summary away a number of the work that isn’t attention-grabbing and doesn’t require a lot thought. In the event you’re not a senior particular person, effectively, then you could find yourself going again to the LLM to ask it to repair its personal errors. And it would try this, however there’s all the time a threat of one other mistake being launched, and on down the street. There’s actually no telling how lengthy this might take, and it’s laborious to check the time with how lengthy it will have taken you to be taught the underlying code and simply remedy it your self.
Nonetheless, a lot of folks will argue that the LLM repeating workflow is okay. Individuals might imagine that simply being there to click on the hover button to inform Copilot to “Repair” time and again is enough, even when it’s not a very satisfying job. Nonetheless, there are severe arguments to be made that the output will likely be significantly flawed, whether or not in readability, maintainability, safety holes, or different features.
“Certain, Gen AI supercharges growth, but it surely additionally supercharges threat. Two engineers can now churn out the identical quantity of insecure, unmaintainable code as 50 engineers.” — https://www.zdnet.com/article/10-professional-developers-on-vibe-codings-true-promise-and-peril/
This query has been addressed by many writers lately. Nonetheless, I’m much less at this second in whether or not Copilot can get to the specified end result finally than I’m in what this course of and approach of working does to us.
Broaden the Aperture
Do not forget that we’re right here to speak in regards to the impact of the LLM on the career. Within the pre-AI period, a junior engineer would begin engaged on a ticket, and make some errors, introduce a bug or two, they usually’d principally be armed with StackOverflow and the docs, plus friends to collaborate with for assist, to determine methods to get it proper. As soon as they muddled by and received the code working, they’d make a PR and get suggestions from extra skilled friends, who would supply suggestions and corrections. As soon as these corrections had been made, a PR might be permitted and merged.
I believe the “muddling by” a part of this course of is definitely pedagogically and professionally fairly essential. When you’ve handed by the educational preparation for the software program engineering position, the sensible preparation is what you could develop the capabilities to develop. I’m a powerful believer in the concept that making errors and fixing them is a tremendously useful option to be taught. As soon as the junior dev engaged on that little bit of code has labored laborious, mounted their very own bugs, written the assessments and docs, and gone by PR evaluation, they know that little bit of code higher than another particular person round. That course of is the productive cognitive labor that brings them a step nearer to being an engineer who can work independently, and tackle extra senior tasks efficiently.
On Being Environment friendly
Nonetheless, from the skin, this course of appears gradual. It appears like that they had to spend so much of time flailing round to only get to the top results of code that works, and absolutely it will be extra environment friendly to both simply make an LLM do it, or make a senior dev do it?
If effectivity is your solely standards, then positive, possibly that is true. However I wish to push again strongly on the premise. We do a lot of issues with out enthusiastic about broader context or impression, particularly within the tech trade, and prioritizing effectivity immediately in any respect prices is a recipe for future catastrophe. Let’s spin out the narrative a bit — if effectivity is the precedence, both we’ve an LLM do the majority of the work, or a senior dev does it, or possibly a little bit of each, like an LLM does the primary stab and a senior dev opinions it. Setting apart any questions of high quality (about which there ought to be questions), what’s the position of a junior SWE on this house?
Does a junior SWE watch the LLM do a lot of the coding? Do they simply exist as an observer? Do they click on “okay” occasionally because the LLM goes alongside, exemplifying “vibe coding”? Or do they not exist in any respect on this state of affairs? Loud voices in the AI space certainly are arguing that they disappear altogether. If that’s the case, within the speedy second, employers in tech save a big amount of cash on salaries, so inventory value goes up.
The concept of vibe coding is utilizing an LLM primarily based software to put in writing the code in your challenge, with minimal precise hand-writing of code concerned. The identify comes from the precept that you simply behind the keyboard have a common sense of the “vibe” of what you need your end result to be, however you both don’t know methods to do it your self or don’t wish to do it your self, so that you’re outsourcing the cognitive work of coding to a pc. You’ll simply describe the broad strokes of what you need, in human language, and let the LLM work out a option to obtain that by code.
Hollowing
Nonetheless, we all know {that a} wholesome career must be bringing in new expertise as senior employees progress and finally retire. Senior engineers are nonetheless being wanted, as a result of they will remedy the issues and write the advanced code that neither junior devs nor LLMs can produce. Their expertise and abilities are undeniably useful and essential to good software program being produced. However the place do we predict senior engineers come from? Identical to senior medical doctors, or senior attorneys, they begin out as junior folks. Their capabilities should be acquired by expertise.
If we construct a working surroundings the place junior software program engineers not exist, we’re hollowing out the pipeline of expertise the career wants. As a substitute of getting a funnel of expertise from educational research into the sphere, younger folks haven’t any actual option to get to the extent of expertise that’s in highest demand.
Conclusions
I admire readers coming together with me on this dialogue, as a result of it’s essential to consider the know-how we make ubiquitous not simply from our personal particular person views however from a broader view. How we make use of LLM know-how within the discipline of software program engineering may have a severe impression on the way forward for the career and what it appears like — and, I’d argue, that is true for a lot of different fields as effectively the place incorporating LLMs is changing into modern. I’m not arguing for zero LLM utilization in coding, by any means — it may be extremely helpful you probably have the expertise and abilities to take advantage of it. (There are numerous essential conversations to have about the way it might impression our talent ranges and the upkeep of our abilities, however that’s a unique subject for an additional day.)
However in any career, senior practitioners don’t simply seem — they’re created, and that technique of creation takes work and time, and will not be essentially the most instantly “environment friendly” factor to do. Effectivity within the speedy second simply isn’t an important factor — it will possibly’t be, if we wish the long run to be vivid.
I’m talking at two occasions in July 2025, together with a free virtual conference called “Agents in Production” — please join in the event you’re !
Learn extra of my work at www.stephaniekirmer.com.
Additional Studying
AI Killed My Job: Tech workers
Tech workers at TikTok, Google, and across the industry share stories about how AI is changing, ruining, or replacing…www.bloodinthemachine.com
10 professional developers on vibe coding’s true promise and peril
Is vibe coding the future of software or a security nightmare in disguise? Here’s what experienced developers think…www.zdnet.com
Frontiers | Some Evidence on the Cognitive Benefits of Learning to Code
Please see the submitted files.www.frontiersin.org
Trusting your own judgement on ‘AI’ is a huge risk
Web dev at the end of the world, from Hveragerði, Icelandwww.baldurbjarnason.com
Jones, Capers. 2014. The Technical and Social History of Software Engineering.
Layoffs.fyi — Tech Layoff Tracker and DOGE Layoff Tracker
LIVE] Tracking all tech startup layoffs — and lists of employees laid off — since COVID-19 was declared a pandemic…layoffs.fyi
