@elliedaw.bsky.social
Mostly Claude Code and Cursor, how about you?
July 14, 2025 at 10:14 PM
Mostly Claude Code and Cursor, how about you?
Same here! The whole workflow is a bit fragile right now - you have to have background knowledge, know how to prompt them, how to split up tasks in a way they understand, etc. but it’s just the early days
July 14, 2025 at 9:44 PM
Same here! The whole workflow is a bit fragile right now - you have to have background knowledge, know how to prompt them, how to split up tasks in a way they understand, etc. but it’s just the early days
AI coding agents won't kill software engineering, in fact we can see a little taste of a future of software where they actually enable engineers to spend most of their time on the fun parts. But for complex codebases and products, we are not close... the context awareness gap has to narrow.
July 14, 2025 at 3:19 PM
AI coding agents won't kill software engineering, in fact we can see a little taste of a future of software where they actually enable engineers to spend most of their time on the fun parts. But for complex codebases and products, we are not close... the context awareness gap has to narrow.
They're burning out because it would just be faster to have done it themselves from scratch... they would have implemented the context-aware solution the first time.
July 14, 2025 at 3:19 PM
They're burning out because it would just be faster to have done it themselves from scratch... they would have implemented the context-aware solution the first time.
This disconnect between the design layer and the implementation layer is what's stopping the AI coding agents from letting engineers focus on the fun parts. Engineers are spending time reviewing and rewriting implementations that don't make sense in their context.
July 14, 2025 at 3:19 PM
This disconnect between the design layer and the implementation layer is what's stopping the AI coding agents from letting engineers focus on the fun parts. Engineers are spending time reviewing and rewriting implementations that don't make sense in their context.
The agents don't have the context of the earlier steps in the workflow. Even when they do, they still don't produce solutions that fit the context.
They generate code like nobody's business, but they don't distinguish between solutions which implement the same feature but suit different contexts.
They generate code like nobody's business, but they don't distinguish between solutions which implement the same feature but suit different contexts.
July 14, 2025 at 3:19 PM
The agents don't have the context of the earlier steps in the workflow. Even when they do, they still don't produce solutions that fit the context.
They generate code like nobody's business, but they don't distinguish between solutions which implement the same feature but suit different contexts.
They generate code like nobody's business, but they don't distinguish between solutions which implement the same feature but suit different contexts.
AI coding agents help with the last step, which is actually awesome because the fun for most engineers is in the other steps.
But in reality, plugging the agents in at the final step isn't resulting in productivity gains for software engineers.
But in reality, plugging the agents in at the final step isn't resulting in productivity gains for software engineers.
July 14, 2025 at 3:19 PM
AI coding agents help with the last step, which is actually awesome because the fun for most engineers is in the other steps.
But in reality, plugging the agents in at the final step isn't resulting in productivity gains for software engineers.
But in reality, plugging the agents in at the final step isn't resulting in productivity gains for software engineers.
The fun part of software engineering has always been in the complexity.
Understanding the problem. Thinking about possible solutions. Weighing the pros and cons. Deciding on the best path. Designing the details. Then finally seeing the solution come together.
Understanding the problem. Thinking about possible solutions. Weighing the pros and cons. Deciding on the best path. Designing the details. Then finally seeing the solution come together.
July 14, 2025 at 3:19 PM
The fun part of software engineering has always been in the complexity.
Understanding the problem. Thinking about possible solutions. Weighing the pros and cons. Deciding on the best path. Designing the details. Then finally seeing the solution come together.
Understanding the problem. Thinking about possible solutions. Weighing the pros and cons. Deciding on the best path. Designing the details. Then finally seeing the solution come together.
My whole career has been in this intersection. It's the raison d'être at gjalla.io. Software engineering isn't going away, and engineers need tools that enable them to do the fun (and challenging) work.
June 27, 2025 at 3:20 PM
My whole career has been in this intersection. It's the raison d'être at gjalla.io. Software engineering isn't going away, and engineers need tools that enable them to do the fun (and challenging) work.
I'm on team engineer! The opportunity of AI coding tools is that engineers could spend more time on the fun stuff. To get there... we need to bridge the context gap: better visibility, better tools for the fun tasks, and a direct connection between the system design and the coding tools.
June 27, 2025 at 3:20 PM
I'm on team engineer! The opportunity of AI coding tools is that engineers could spend more time on the fun stuff. To get there... we need to bridge the context gap: better visibility, better tools for the fun tasks, and a direct connection between the system design and the coding tools.
So engineers are spending more time reviewing massive PRs and rewriting huge chunks of code that were generated with a poor understanding of the context of the system... and spending less time on the fun stuff: thinking through problems, weighing tradeoffs, and figuring out the best solution.
June 27, 2025 at 3:20 PM
So engineers are spending more time reviewing massive PRs and rewriting huge chunks of code that were generated with a poor understanding of the context of the system... and spending less time on the fun stuff: thinking through problems, weighing tradeoffs, and figuring out the best solution.
Why? The tools are good for discrete tasks, but aren't there yet for complex codebases and production-ready code. They add chaos to the codebase and teams lose visibility into what's happening.
June 27, 2025 at 3:20 PM
Why? The tools are good for discrete tasks, but aren't there yet for complex codebases and production-ready code. They add chaos to the codebase and teams lose visibility into what's happening.
AI coding tools in particular are helping a lot of people move faster, but at the same time it's causing burnout and low morale for senior engineers.
June 27, 2025 at 3:20 PM
AI coding tools in particular are helping a lot of people move faster, but at the same time it's causing burnout and low morale for senior engineers.
Being awesome at writing code and being awesome at building software are not the same thing... enter gjalla.io ✨ if you've felt this pain too and want to be a part of the beta, please DM me... I want to talk to you!
June 26, 2025 at 2:15 AM
Being awesome at writing code and being awesome at building software are not the same thing... enter gjalla.io ✨ if you've felt this pain too and want to be a part of the beta, please DM me... I want to talk to you!
I've been playing with our beta for a couple of days and... y'all it's really cool. I had tried refactoring this one component 3 times in the past and had to revert every time. The agents kept getting into unfixable error loops. Hooked up the gjalla MCP and I got through the refactor in one evening.
June 26, 2025 at 2:15 AM
I've been playing with our beta for a couple of days and... y'all it's really cool. I had tried refactoring this one component 3 times in the past and had to revert every time. The agents kept getting into unfixable error loops. Hooked up the gjalla MCP and I got through the refactor in one evening.
how many times have you made a decision about how to architect something and then the agent goes off and does the opposite thing anyway? or it implements an entire feature but in a way that doesn't match how similar features are already implemented in your codebase...???
June 26, 2025 at 2:15 AM
how many times have you made a decision about how to architect something and then the agent goes off and does the opposite thing anyway? or it implements an entire feature but in a way that doesn't match how similar features are already implemented in your codebase...???
I'm seeing lot's of engineers that are feeling burned out lately... there's so much opportunity for better tooling at the fun layers (this is the gap we're working on at gjalla.io 👀)
June 23, 2025 at 4:49 PM
I'm seeing lot's of engineers that are feeling burned out lately... there's so much opportunity for better tooling at the fun layers (this is the gap we're working on at gjalla.io 👀)
- ...maybe even the ability to implement multiple design decisions tailored to use-cases that can absorb the respective tradeoffs?
It's exciting. These are the problems that engineers have fun solving.
It's exciting. These are the problems that engineers have fun solving.
June 23, 2025 at 4:49 PM
- ...maybe even the ability to implement multiple design decisions tailored to use-cases that can absorb the respective tradeoffs?
It's exciting. These are the problems that engineers have fun solving.
It's exciting. These are the problems that engineers have fun solving.
- We can spend less time in architecture reviews talking about potential implications and instead review with data specific to our context (because we can build and test it more than before!)
- We'll have a much better ability to manage the tradeoffs because we can see them in action faster
- We'll have a much better ability to manage the tradeoffs because we can see them in action faster
June 23, 2025 at 4:49 PM
- We can spend less time in architecture reviews talking about potential implications and instead review with data specific to our context (because we can build and test it more than before!)
- We'll have a much better ability to manage the tradeoffs because we can see them in action faster
- We'll have a much better ability to manage the tradeoffs because we can see them in action faster
The "once well-defined" piece is where it gets interesting... with the cost of writing code getting lower, we're moving toward having more space to actually test multiple approaches.
If the trend continues...
If the trend continues...
June 23, 2025 at 4:49 PM
The "once well-defined" piece is where it gets interesting... with the cost of writing code getting lower, we're moving toward having more space to actually test multiple approaches.
If the trend continues...
If the trend continues...
This isn't just a communication problem, it's a strategic one, and a competitive advantage for those who solve it.
May 6, 2025 at 3:01 PM
This isn't just a communication problem, it's a strategic one, and a competitive advantage for those who solve it.