Yes and No.

  Yes:  Agents will be there as an option, with less SWEs needed.
   No:  Agents won't be a complete SWE replacement and they still need human SWEs for accountability.

Here's a great analogy on driving with autopilot:

Say you are driving your car on autopilot. What happens when it stops working or experiences an outage / malfunction?

Do you sit there and wait for the provider to get back online or do you take control of the wheel yourself?

So having said that:

> All code reviews are agent driven. No one takes the time to actually understand anything. Documentation has become novel length slop, as have Jira tickets.

> I ship stuff I don't understand.

Looking at my above sentence and judging by this workflow, is the future of driving having people never looking on the roads and no hands on the wheel while driving and they should wait for the provider to fix the outage whilst being stranded on the motorway?

Both driving and SWE will always require a human in the loop in case the system fails and requires human intervention.

Sounds fucking miserable. Thank goodness I'm far enough into my career that I can throw in the towel and retire instead of having to capitulate to this lunacy.

You are not alone, but it is not the future. I currently am working a short-term gig within a company with a similar setup - the leadership wants to go that way, and some teams are doing so. Those teams are also failing. Their code passes tests, but doesn't hold up to real usage. The team members don't know the code well enough to even talk about the problems. The DevOps teams don't know the platforms well enough to talk about problems either. Some of the most senior folks in the org will just paste AI slop into a document, then not even be able to navigate their own document on a call to explain things.

"Miserable" is absolutely the right word for it. "Dystopian" is another word I use to describe it. Either way, the results I'm seeing make me think that this trend won't last. These companies will either figure out that it isn't working, or they will fail.

FWIW, "AI-first" is one thing. "AI-only" is where it crosses from a strategic choice into a f'ing pile of stupidity.

How does your team handle security audits or compliance? If the code reviews are also agent driven, you have created a closed loop system with no human accountability.

You are definitely not alone.

I worked at a place which took a similar approach. As someone whose been doing this for 30 years, I found this to be chaotic, wasteful and antithetical to everything I've learned about software development or even basic communication. This was months ago and things are moving fast. Lately, I see best practices starting to emerge and have confirmed with friends working at other companies that people are employing them.

Is this the future though?

The sense that I get is that a lot of us are still trying to figure things out. Giants like Martin Fowler and Kent Beck have said as much. This article written by Annie Vella sums it up quite well:

https://annievella.com/posts/finding-comfort-in-the-uncertai...

I periodically check in with people I have worked with in the past to see what they are up to and everyone is in a similar boat in terms of trying to figure it all out. Honestly, it feels a little like we are all at one of the different stages of grief. Some are angry and in complete denial. Others are absolutely delusional about the purpose and capabilities of this technology. Most of us I suspect are simply trying to process what is happening.

No. We have one team at our company doing this. Management loves them, but no one in engineering trusts them and they literally just write MCPs that have no users and still experience daily incidents.

Is the software produced this way good quality?

No, the current state of things isn't stable enough to last a long time. I don't know what the future will look like but it can't be this, and it's not going to be rolling the clock back 5 years either.

You’re not alone and I really hope this isn’t the future of work. It’s unsustainable to push out code you dont understand. It’s one thing to prototype, or build internal tools, but this cannot scale to commercial offerings (for much longer). With agents gone rouge in the news eroding trust. Just my 2 cents.

I think human in the loop will still be a thing for critical workloads for a while, even if it’s agent-first and human pulled as escalations are needed.