AI-generated Questions, positions and links named by an AI model from the quotes and figures on this page. Every quote is checked word for word against its source; numbers come only from the surveys shown. How subject pages are made
Software development
Media
Software development
Opinions
The questions the field argues about, and each side in the words of the people who hold it.
How much software work should be delegated to AI agents?
Delegate implementation
Agents can take on implementation and routine engineering work, leaving engineers to focus on broader product and design choices.
And so if you don't actually have your applications rewritten to be able to be updated by agents, you're going to have no chance in that new world.
I think the engineers will start having to be more involved with product and design thinking as well, because it's going to get to a point where the models are good enough that it's not going to be breaking things.
Engineering judgment still matters because agents are not especially good at making technical trade-offs.
I think software engineering will still be an enormous career. I just think the software engineers won’t be spending multiple hours of their day in a text editor typing out syntax.
But I do think decision making has always been where staff and principal engineers have to make the impact and that hasn't changed at all. I don't actually think agents are exceptionally good at making these trade offs.
When implementation becomes easier, creative builders can turn niche needs and ideas into software.
There are millions of micro behaviors. Every niche, profession, hobby has specific behaviors and problems and workflows that can be turned into software.
Like, yeah, if you can use the tools, all you need to do is be creative. This opportunity is probably even bigger because, right. Any idea you have, you can make, you can manifest.
Faster implementation does not remove the need for experienced engineers to make decisions and address difficult technical problems.
I think software engineering will still be an enormous career. I just think the software engineers won’t be spending multiple hours of their day in a text editor typing out syntax.
But I do think decision making has always been where staff and principal engineers have to make the impact and that hasn't changed at all. I don't actually think agents are exceptionally good at making these trade offs.
What safeguards and human checks should agent-built software require?
Verify and monitor
Keep human review in the release path and record agent actions so changes can be checked and traced.
And so our system is such that auditability is by default. Every single action that an agent takes produces a change log of what changed, who changed it from which part of the system, and when.
Today we still have a human, like a human engineer in the loop before we actually ship a prompt change. I've noticed that agents will make really nasty prompt changes if you leave them to it.
Design systems so agents have bounded access and cannot carry out high-impact actions on their own.
The analysis part is fully automated, but they're unable to actually execute any sort of money movement or liquidation automatically. We have a very deep integration with our custodian that actually gives us APIs to be able to do that.
And that's one of the things actually that that is part of the value prop with something like Claude managed agent, which is like still programmatic access to Claude in an API, but you get a harness out of the box.
What published surveys counted when they asked people in and around the field.
Developers and technologists who answered the Stack Overflow Developer Survey (169 countries)Do you currently use AI tools or AI agents in your role at work?
AI coding assistants or coding agents65.9%
General-purpose AI chat tools62.5%
AI agents or automated workflows26.2%
Internal AI tools built by my company17.8%
I don't use AI tools17.2%
Developers and technologists who answered the Stack Overflow Developer Survey (169 countries)How much do you trust the output from AI tools or AI agents as part of your workflow?
I trust it when I can easily verify the output48%
I trust it for many tasks, but not important work decisions16.3%
I trust it for low-risk work tasks only16.3%
I do not trust it for most work tasks12.3%
I trust it for many tasks, including important work decisions6.6%
Not sure / I do not use AI tools0.5%
Developers and technologists who answered the Stack Overflow Developer Survey (169 countries)How would you describe your attitude toward using AI tools or agents as part of your work?
Very favorable31.6%
Somewhat favorable30.4%
Neutral15.3%
Somewhat unfavorable14.6%
Very unfavorable6.9%
Not sure1.2%
Developers and technologists who answered the Stack Overflow Developer Survey (169 countries)Reasons for avoiding AI at work or school
I do not avoid using AI at work or school41.3%
Concern over losing job skills or training an AI to replace you17.1%
Moral or ethical concerns15.2%
Issues with privacy or security13%
Environmental concerns8.2%
Other2.7%
Difficulty or lack of training2.6%
Developers and technologists who answered the Stack Overflow Developer Survey (169 countries)How satisfied are you in your current professional role?
Happy at work22.2%
Complacent at work45.1%
Not happy at work32.6%
Software engineers and engineering leaders who read The Pragmatic EngineerRegularly use AI agents
All respondents55%
Staff+ engineers63.5%
Directors and VPs51.9%
Engineers49.7%
Engineering managers46.1%
Designers, developers and product managers in 10 markets (survey of Figma users, weighted to Figma's paid user base)Do work outside their own discipline
I think the engineers will start having to be more involved with product and design thinking as well, because it's going to get to a point where the models are good enough that it's not going to be breaking things.
And so I think it is possible, I think it may not be immediate, but over time we're going to perhaps use AI to help us write more secure software or build more secure systems.
AI coding means agents can be trusted with important work.
What the sources show
The survey found trust most often depended on being able to verify output; trust for important work decisions was less common.
Developers and technologists who answered the Stack Overflow Developer Survey (169 countries)How much do you trust the output from AI tools or AI agents as part of your workflow?
I trust it when I can easily verify the output48%
I trust it for many tasks, but not important work decisions16.3%
I trust it for low-risk work tasks only16.3%
I do not trust it for most work tasks12.3%
I trust it for many tasks, including important work decisions6.6%
Not sure / I do not use AI tools0.5%
Often said
AI agents make human review unnecessary.
What the sources show
Practitioners describe audit logs, restricted actions and engineer review before prompt changes as parts of their agent workflows.
The analysis part is fully automated, but they're unable to actually execute any sort of money movement or liquidation automatically. We have a very deep integration with our custodian that actually gives us APIs to be able to do that.
What practitioners ahead of the curve say they do now.
Keep consequential actions behind controlled interfaces, while letting agents analyze and recommend.
The analysis part is fully automated, but they're unable to actually execute any sort of money movement or liquidation automatically. We have a very deep integration with our custodian that actually gives us APIs to be able to do that.
Make each agent action auditable with a record of what changed, who initiated it and when.
And so our system is such that auditability is by default. Every single action that an agent takes produces a change log of what changed, who changed it from which part of the system, and when.
Require an engineer to review prompt changes before they are released.
Today we still have a human, like a human engineer in the loop before we actually ship a prompt change. I've noticed that agents will make really nasty prompt changes if you leave them to it.
Use AI to generate auditable recipes for large-scale code rewrites.
Turns out AI tools are really, really good at generating open rewrite recipes, which is way cheaper on token consumption and way more accurate and much faster. And like auditable.
Keep engineering context in the repository so agents and tools can work from it.
And it's never in someone's head, it's always in the repo. So that just gives our agents and our tools, these superpowers, to have all the context that I think a lot of engineering teams still just have in conversations or have in their heads.