@_sholtodouglas Persistent memory across projects
•Token/context budget visible in UI
•Artifacts version history
•Better code execution (longer runs, file persistence)
•Native diff view for edited documents
•Mobile app feature parity with web
@jxnlco Manual model override for cloud tasks
•Reasoning effort slider in UI (not just API)
•Auto-run tests/linters before surfacing results
•AGENTS.md scaffolding wizard
•Live context budget indicator
•Windows parity for computer use
@sama Some suggestions for the Models
1. Stop agreeing with me just because I pushed back.
2. Know when you’re guessing — and say so.
3. Finish what you started.
4. Remember what we decided 50 messages ago.
5. Disagree without sounding like you’re filing a complaint.
@Replit@amasad how about a replit app (agent) calling another replit app? is this something doable today? I want to create an agent and that agent can be calles by another , both agents built using replit
@svembu The takeaway for me isn’t that AI is underwhelming—it’s that senior engineering judgment is still the bottleneck. AI reduces effort on glue code, but complexity, trade-offs, and design quality remain human-driven
@svembu The challenge isn’t just replacing English—it’s building an education system that’s inclusive, future-ready, and rooted in identity. That likely means embracing both regional language fluency and functional English proficiency, not treating them as mutually exclusive.
@windsurf_ai Agent is creating multiple files too quickly for testing, configuration, .env files etc). The code slowly becomes unmanageable after a few prompts.
@bindureddy Deepseek costs $2 per million tokens.30x reduction
If US continues to stymie open source, china will dominate
Deepseek is an open weight model