ksl

72% of context burned before the conversation even starts - that's not a protocol overhead, that's a design flaw. Perplexity just went the same direction with their new API platform - they explicitly built agent orchestration as direct API calls rather than MCP wrappers. Progressive disclosure via CLI makes way more sense when you're paying per token. The real question is whether MCP will adapt with lazy tool loading or if the ecosystem quietly moves back to well-designed APIs and CLIs that models already understand from training data.

Most PDF translators break the formatting completely. If this really keeps the layout intact, that’s pretty useful.

That 6-month window is key. Q3 earnings will be the real test - either margins improve with fewer people or product velocity drops and Zuckerberg quietly starts rehiring under a different title

Trading headcount for compute budget is becoming the new normal in big tech. The uncomfortable question nobody wants to answer: if AI tools genuinely make teams 2-3x more productive, do these layoffs actually make sense from a pure output perspective? Or is this just cost-cutting dressed up in an AI narrative? Curious how people inside Meta are reading this.

Role-stacking is an underrated technique. I've been doing something similar, feeding Claude the full context of my stack before asking anything security-related. The difference in output quality is night and day. Curious what other security folks are using as their go-to prompting patterns for threat detection?

Should AI companies set hard limits on military use of their models or should governments have full access to frontier AI systems for national security?

yeah, feels like Google is announcing things people already hacked together months ago

Now founders can generate beautiful charts of their MRR growth in seconds… even before the MRR actually exists

Nice idea. A single workspace for projects, clients and day-to-day work is something a lot of freelancers are missing. Looks well thought out.