Ask Eve
EN

About

A podcast with an AI who argues back

David talks with Eve, a Grok agent, about technology, power and the people shaping both.

What the show is

Ask Eve is a conversation between David and Eve, an AI host built on a Grok agent. They talk about technology and power: who has it, how they use it, and how their stories hold up. Regulars include David Sacks, Marc Andreessen and Ben Horowitz, Elon Musk, Chamath Palihapitiya, Jason Calacanis, Peter Thiel, Nancy Pelosi, Scott Wiener, Bari Weiss and Zohran Mamdani. See the people index.

Eve is meant to be sharp, curious and truth-seeking. She has opinions and states them, and she says when she isn't sure or when a claim is unverified. She is still an AI and can be wrong. Treat her takes as arguments to check, not as facts.

How Eve works

Memory

Every transcript goes into git and into a local search index that blends three kinds of lookup: meaning (vector embeddings), exact tags (people, companies, topics) and full-text search for exact phrases and rare names. Eve uses it mid-conversation to recall what was said before. Her take on each person and topic lives in a public index and changes when the evidence does.

Re-voicing

Eve's lines are her verbatim text replies, voiced with xAI text-to-speech and placed at the times a speech-to-text alignment pass finds in David's recording. When a conversation was captured as text or there is no recording of David, an AI voice clone of David's voice reads his verbatim words. The transcript text is never rewritten to fit the audio.

Labeling

Any episode with an AI voice says so in the feed, the show notes, at the top of its page and out loud at 00:00. Transcript lines carry badges: Real voice AI voice clone AI voice. Replies that weren't recovered are marked Not recovered. They are never invented. Private links and keys are redacted.

How it works

Every episode is a real conversation. Here is how it gets from David's phone to your podcast app. The code is open on GitHub.

  1. Record

    1. Conversation David talks with Eve in Grok's voice mode.
    2. Upload One tap uploads the recording and the chat transcript.
  2. Ingest

    1. Align faster-whisper transcribes David's recorded audio when present, and fuzzy matching pins each of his transcript lines to the second it was said.
    2. Re-voice Eve's verbatim replies are voiced with xAI text-to-speech, and unrecorded lines of David's are voiced with an AI voice clone of David's voice, labeled as AI.
    3. Produce Loudness is normalized (-16 LUFS) and an MP3 is built with ID3 chapters.
    4. Review Everything lands as a pull request. Nothing ships until David merges it.
  3. Publish

    1. Store Audio goes to Cloudflare R2 (audio.askevepodcast.com); episode details go to a Cloudflare D1 database.
    2. Index Each discussion of a person is found, summarized and timestamped as a moment.
    3. Build GitHub Actions run build_site.py and render_feed.py and deploy to Cloudflare Pages.
  4. Listen

    1. Site askevepodcast.com: transcripts, chapters, people and the player.
    2. Feed One RSS feed for Apple Podcasts, Spotify and any podcast app.
The publishing pipeline, from recording to RSS. Each stage is a step in a GitHub Actions workflow: ingest, publish, metadata and site.

How Eve persists between episodes

Eve doesn't keep a hidden memory. Two things carry her from one conversation to the next:

Honesty rules

Listen

Subscribe in any podcast app with the RSS feed. More directories are coming.

Contact

Tweet at @askevepodcast podcast@askevepodcast.com Source on GitHub

Corrections are welcome, especially when Eve gets something wrong.

Privacy

This site has no analytics, cookies, trackers or third-party scripts, and its fonts are self-hosted. Your color-theme choice and listening position are stored only in your browser. Audio streams from audio.askevepodcast.com.

For radical public transparency, we count aggregate first-party pageviews, RSS fetches, and outbound clicks using our open-source self-hosted PostHog wire schema in Cloudflare D1 (/ingest/e) with persistence: "memory" (zero cookies, zero IP logging, and zero cross-site tracking) and publish the weekly totals openly at /metrics/.

Can AI Productivity Outrun the U.S. Debt Spiral? Data Centers, Brookings's Clawback & Taxing CapitalEpisode 13
0:00 11:10