Frequently Asked Questions

The Impact of AI-Generated Code on Engineering Workflows

What does "AI slop" mean in the context of engineering?

"AI slop" refers to the influx of AI-generated pull requests (PRs) that often lack context, quality, or alignment with business requirements. Engineers are increasingly tasked with reviewing large volumes of code produced by AI systems, which can lead to frustration and review fatigue when the output is not up to standard.

Why are engineers experiencing review fatigue with AI-generated code?

Engineers are experiencing review fatigue because AI-generated code is often submitted in high volume and lacks the implicit context shared among team members. This results in engineers spending more time reviewing and correcting code, rather than focusing on higher-level design and architecture. The current workflow forces engineers to "babysit" context-free PRs, highlighting a need for new processes and better automated tooling.

How is the role of engineers changing due to AI integration?

The integration of AI is shifting the engineer's role from manual code creation to system design, architecture, and validation. Engineers are increasingly responsible for steering AI towards expected results, ensuring reliability, scalability, and business alignment. As AI matures, engineers will focus more on high-leverage decisions, technical judgment, and accountability rather than line-by-line coding.

What are the main challenges during the transition to AI-assisted engineering?

The main challenges include a lack of trust in AI-generated code, insufficient context for AI systems, and outdated review workflows designed for human-written code. Engineers must actively guide AI output, and teams are experimenting with new processes to establish trust and quality without creating bottlenecks. The transition phase is marked by frustration and inefficiency, but is expected to improve as methodologies evolve.

How are engineering review processes expected to evolve with AI?

Engineering review processes are expected to evolve by moving planning and design upstream, encoding standards and architectural context into playbooks and spec files, and demanding comprehensive test coverage. PR reviews will become AI-assisted, evaluating design, approach, and implementation together. These changes aim to compress review cycles and improve code quality, reducing the bottleneck caused by traditional workflows.

Will the current frustration with AI-generated code reviews last?

The current frustration is considered a transitional problem. As engineering teams adapt their workflows and AI systems gain more context, review fatigue and bottlenecks are expected to decrease. The future engineer will focus on solution building, system orchestration, and high-level judgment, rather than manual code review.

How does Sedai's engineering leadership view the future of engineering in the age of AI?

Sedai's engineering leaders believe that engineering will become more about judgment, context, and accountability. While AI will handle routine code generation, engineers will focus on system design, requirements definition, and ensuring quality across automated workflows. The value of technical judgment and architectural thinking will increase as teams adapt to new tools and processes.

KO THE CLOUD TO ENTER OUR GIVEAWAY

WIN KEYBOARD!

Is AI Slop Ruining Engineering?

Is AI Slop Ruining Engineering?

This week, developers are drowning in AI-generated PRs, and they’re sick of it. One engineer put it bluntly: what's my purpose as an engineer if I'm just reviewing AI slop?

I asked our engineering leadership team what's actually going on. Here's what they said.

Every Industrial Revolution Has a Messy Transition Phase

Hari Chandrasekhar (SVP of Engineering, Core)

The integration of AI is undeniably shifting the nature of software engineering, offloading the routine work of writing code to AI systems. Consequently, the developer's role is elevating from mere code creation to high-level system design, focusing on architecture, framework resilience, and scalability.

This is a natural technological progression similar to the Industrial Revolution's shift from hand-weaving to the mechanized power loom. Early textile workers went from manually threading every single line to operating and troubleshooting complex machines. 

While those early machines often tangled or snapped threads, requiring intense and frustrating oversight, the worker's role ultimately and permanently evolved from manual labor to system orchestration.

That said, the engineer's responsibility hasn't fully transferred. AI is groundbreaking, but a great deal of context in the development cycle is implicit knowledge shared among product managers, analysts, designers, and developers that may not be obvious to an AI orchestrating the work. 

Until that context gap is closed, there will be a transition period where engineers must actively steer AI towards expected results; bad output and slowness will cause real frustration in the meantime. The more context the system has, the better the results.

"There will be a transition period where engineers must actively steer AI towards expected results; bad output and slowness will cause real frustration in the meantime."

Hari Chandrasekhar

Hari Chandrasekhar

SVP of Engineering, Core

This growing pain is very similar to the automotive transition from manual to automatic transmissions. Early drivers of automatic cars often felt the vehicle wasn't shifting efficiently, and constantly wanted to manually override the gears. At the time, this instinct was probably right. 

However, as automotive engineering improved, drivers stopped caring about the internal mechanics of the transmission. They just wanted a smooth, efficient drive. 

As AI systems mature and context windows deepen, the granular, line-by-line internal workings of the generated code won't matter nearly as much. The ultimate measure of an engineer's success will simply be whether the end product is reliable, scalable, and built to meet the needs of the business.

AI Isn’t an Engineer, and We Need To Stop Treating It Like One

Nikhil Gopinath Kurup (SVP of Engineering, ML)

The real issue underlying the AI code bottleneck is trust. We review code because we don't trust it. We don't trust other humans completely, and we certainly don't trust AI to grasp architecture, design principles, or business requirements.

For instance, I wouldn't feel the need to review code written by Fabrice Bellard (nor would I feel qualified to do so) because the trust in his capability is absolute. Similarly, we never review the binary code a compiler outputs because we trust the tool to faithfully and unambiguously translate our intent.

Our current mistake is treating AI as the equivalent of a human developer and forcing it into standard peer-review processes. AI code has distinct merits and severe weaknesses. Forcing engineers to babysit an endless stream of context-free PRs just shows that our current workflow is broken.

"AI code has distinct merits and severe weaknesses. Forcing engineers to babysit an endless stream of context-free PRs just shows that our current workflow is broken."

Nikhil Gopinath Kurup Headshot

Nikhil Gopinath Kurup

SVP of Engineering, ML

The solution isn't to force AI into our old ways of working. We need to build entirely new engineering processes, better automated tooling, and a fundamentally new way of delivering software. The goal is to evolve our methodology until AI output becomes as predictable, reliable, and trusted as a compiler.

Once that workflow is established, the role of the engineer moves one level higher. Instead of doing the grunt work of writing code and checking syntax, developers will focus entirely on design, specifications, and architecture.

Personally, I hate that this removes the raw creativity and fun of writing code line-by-line. But practically, it is the only way forward. If developers are trapped reviewing AI slop, we are stuck in a poorly managed transition phase. The future engineer isn't a coder; they are a solution builder who owns the pipeline from the starting problem to the working product.

Review Fatigue Sucks, but It Won’t Last Forever

Benjamin Thomas (Co-Founder & CTO)

Yes, developers are spending more time reviewing code right now, but I think the current bottleneck is a transitional problem, not a permanent one. A few shifts are already underway that will shorten that cycle significantly.

First, planning and design will move upstream, similar to how traditional design reviews worked — where AI-generated approaches get vetted early, before a single PR is ever opened. 

Second, senior engineers will start building their own skills, playbooks, and spec files that encode their standards and architectural context. This will make reviews progressively faster over time. 

"PR reviews will evolve to be AI-assisted, evaluating design, approach, and implementation together, rather than treating code as the only artifact that matters."

Benji Thomas Headshot (Square)

Benjamin Thomas

Co-Founder & CTO

Third, teams can now realistically demand 100% unit and integration test coverage for far more scenarios than was ever practical before. That alone changes the quality bar at the point of submission. 

And finally, PR reviews themselves will evolve to be AI-assisted, evaluating design, approach, and implementation together, rather than treating code as the only artifact that matters.

Each of these shifts individually helps. Together, they compress the review cycle in a meaningful way. The frustration people are feeling today is real, but it's a symptom of old processes meeting new tools. It’s not a signal that the tools don't work.

Engineering Is Becoming About Judgement, Not Writing Code

Shankar Jothi (VP of Engineering, ML)

In the near term, I expect review and judgment to become even more important engineering skills. The bottleneck is moving from code creation to code validation. That doesn't necessarily mean developers become full-time reviewers, but it does mean the value of context, system knowledge, architectural thinking, and accountability increases. 

AI can help produce code quickly, but someone still needs to understand how that code fits into a larger system and whether it should exist in the first place.

I also think we're still in a transitional period. Many teams are experimenting with AI-assisted development, but the surrounding processes haven't fully adapted yet. 

"Developers won’t become full-time reviewers, but the value of context, system knowledge, architectural thinking, and accountability is increasing."

Shankar Jothi Headshot

Shankar Jothi

VP of Engineering, ML

Code review workflows, testing practices, ownership, and engineering standards were largely designed around humans writing code. As AI becomes a larger contributor, teams will likely need new ways to establish trust and quality without creating review bottlenecks that slow everyone down.

For me, the role of the developer is not becoming less important; it's becoming more focused on higher-leverage decisions. Writing code will remain part of the job, but understanding systems, defining requirements, evaluating trade-offs, and ensuring quality across increasingly automated workflows may become an even larger part of engineering. 

The tools are changing rapidly, but the need for technical judgment and accountability remains constant.

Code Quality Is Collapsing in the Age of Infinite PRs

Suresh Mathew (CEO)

The real problem isn't that developers are becoming reviewers, it’s that the bar for code submissions has collapsed. When generating a PR costs nothing, and everyone submits, the quality of our code disintegrates. 

And there's no going back. METR's follow-up study couldn't get clean data on whether AI actually makes developers more productive — no one would work without it long enough to form a control group. 

The dependency on AI is structural now, and moving forward means figuring out how to keep reviews rigorous without becoming the bottleneck.


The way your team reviews and ships code is changing fast. See how Sedai helps engineering teams keep up.

About the Author

Suresh Mathew

Founder & CEO

Suresh Mathew is the CEO & Founder of Sedai. A pioneer of autonomous cloud optimization, Suresh works with Fortune 500 companies and government agencies to optimize their cloud & AI infrastructure. Under Suresh’s leadership, Sedai has expanded to a team of 100+ employees around the globe, received eight U.S. patents on its technology, and secured $38 million in funding from top investment firms.

Read my Full Bio

Similar Resources

Why Won't AI Simplify Code?

How Do We Deal With the Complexity AI Agents Hand Us?

AI agents ship overcomplicated code and argue when asked to simplify it. Three Sedai engineering leaders on how they review, budget, and cut it down.

Eng Leaders React Meta's Agent

Does Meta Seriously Expect Us to Hand Over Our Code?

Meta's Muse Code offers cheap tokens if you let Meta train on your code. Three Sedai engineering leaders on why that trade is a non-starter for product code.

Eng leaders react ai mania is evicerating global decision making

Is AI Too Big to Question?

A viral essay claims a 0% AI project success rate and dissent gets people fired. Three Sedai engineering leaders on how to question AI with numbers.

Would You Ship Your Codebase ReWritten By Agents?

Would You Ship a Codebase Rewritten Entirely by Agents?

Bun ported 535,000 lines of Zig to Rust with 64 Claude agents in 11 days. Four Sedai engineering leaders on when agent-written code is safe to ship.

eng-leaders-react-how-much-isolation-ai-agents-need-featured-image-blog

How Much Isolation Do AI Agents Actually Need?

AWS, Microsoft, Google, and Anthropic all isolate AI agent sessions differently. Sedai's engineering leaders weigh in on how much isolation agents really need.

No More Tokenmaxxing Notes From Suresh

Tokenmaxxing Is Ending. The Era of AI Tokenomics Has Begun

Token consumption was never the right metric. Here's why enterprise AI budgets are exploding and what to measure instead.