the future workflow is likely:
devs define architecture, product intent, and behavioral constraints -> AI writes code + AI writes tests + AI runs the tests + AI validates
agree?
the future workflow is likely:
devs define architecture, product intent, and behavioral constraints -> AI writes code + AI writes tests + AI runs the tests + AI validates
agree?
I keept forgetting shell syntax, built an AI powered Linux CLI helper:
ai -x "find all .log files larger than 100MB"
- uses local Ollama models
- reads recent shell history (context)
- suggests commands before execution
keeps me inside the terminal instead of looking up syntax
I keept forgetting shell syntax, built an AI powered Linux CLI helper:
ai -x "find all .log files larger than 100MB"
- uses local Ollama models
- reads recent shell history (context)
- suggests commands before execution
keeps me inside the terminal instead of looking up syntax
DESIGN SYSTEM is the one of the most important parts of software development you shouldn't skip!
it defines:
- colors
- typography
- spacing
- components
- interaction rules
- accessibility
- usage guidelines
without it, AI agents generate inconsistent UI chaos...
DESIGN SYSTEM is the one of the most important parts of software development you shouldn't skip!
it defines:
- colors
- typography
- spacing
- components
- interaction rules
- accessibility
- usage guidelines
without it, AI agents generate inconsistent UI chaos...
Why do users pay for AI coding agentsโ (like Claude Code) mistakes?
If the agent gets stuck in loops, rewrites tests 5 times, or burns massive context windows, the bill still goes to the user.
Whereโs the accountability for token waste?
Tokens are officially the new gold... so how not to burn through quotas?
One thing I keep seeing with ClaudeCode (and I guess same happens in other agents) - they repeatedly generate & rewrite tests just to validate code
Sometimes they even fix bugs in the tests themselves.. =)
Tokens are officially the new gold... so how not to burn through quotas?
One thing I keep seeing with ClaudeCode (and I guess same happens in other agents) - they repeatedly generate & rewrite tests just to validate code
Sometimes they even fix bugs in the tests themselves.. =)
Why not generate reusable unit / smoke / ... tests once, save them? The agnet would just rerun them every think it need to verify the code...
Would save tons of tokens.
Tokens are officially the new gold... so how not to burn through quotas?
One thing I keep seeing with ClaudeCode (and I guess same happens in other agents) - they repeatedly generate & rewrite tests just to validate code
Sometimes they even fix bugs in the tests themselves.. =)
wellโฆ that happened ๐
asked an AI to build an agent and it basically said - maybe donโt...
what do you even do when the AI tells you NOT to make it fully autonomous?
Unpopular opinion: many "AI agents" can be replaced with simple automation...
You should use AI only where itโs actually needed... not just because itโs hype =)
Doing this can drastically reduce costs.
With all the 'AI slop', GIT is the only tool actually gaining signal.
- are you using GIT?
or just overwriting files and hoping for the best? =D
be honest... =)
agents are valuable when:
- The task requires reasoning
- Inputs are unstructured / unpredictable
- Rules are hard to define upfront
(these are only a few examples, there are many more in practice)
Unpopular opinion: many "AI agents" can be replaced with simple automation...
You should use AI only where itโs actually needed... not just because itโs hype =)
Doing this can drastically reduce costs.