"I used to spend an hour just organizing notes after listening to a podcast episode I wanted to write about. Now I drop a link into Verdent and 20 minutes later I have content for 4 platforms."
— Ethan Lim, Developer Advocate, Singapore
Ethan Lim is a Developer Advocate. He listens to a lot of developer podcasts — on the MRT, between meetings, while making coffee — and he's constantly hearing ideas he wants to share. The problem was always what to do with them.
A 90-minute interview isn't something you can ask other people to sit through. So if any of those ideas were going to reach anyone else, Ethan had to do the writing. His old routine: re-listen at 2x (still 45 minutes), jot notes in a separate app, then write three different versions — one for Twitter, one for LinkedIn, one for his blog. Three to four hours per episode. More than half the time, he'd abandon it halfway.
Now he drops the podcast URL into a Verdent project he set up called PodForge. Two minutes later he has a clean transcript. Five minutes after that he's skimmed it, marked three or four moments worth talking about, and told Verdent what he actually thinks. Ten minutes later there's a Twitter thread, a LinkedIn article, an Instagram carousel (with PNG slides exported and ready), and a long-form blog post — each shaped for its platform, each with timestamped quotes, each carrying his own point of view. One podcast link. Under 20 minutes, end to end.
A typical session
Ethan drops the URL straight into Verdent:
/transcribe https://podcasts.apple.com/podcast/id123456/episode-slugtranscribe is a skill Ethan wrote himself. It tells Verdent how to handle the whole pipeline: figure out the platform, download the audio, pick a transcription engine, and drop three files into workspace/transcripts/ — the original audio, a timestamped SRT, and a plain-text transcript. About two minutes. When it's done, Verdent gives him a short summary and asks whether he wants the transcript cleaned up.
He almost always says yes. And he's learned that cleanup quality is directly proportional to how specific the context is — so he built his transcribe skill to have Verdent fetch the original webpage, which often contains key background info like guest names, topic keywords, and technical terms mentioned in the discussion. Verdent then fixes speaker names, corrects technical terms (instead of mangling "gRPC" into something unreadable), strips filler ("um," "you know"), and breaks walls of text into actual paragraphs. The cleaned version goes into workspace/cleaned/, and any subsequent /write calls use that version directly.
Now comes the part only Ethan can do.
He doesn't read the transcript word-for-word — he skims. What he's looking for is three to five spots that make him think this is worth saying, and he jots down the rough timestamps. Then he figures out his own angle: what he agrees with, what he doesn't, what he'd add. This step matters. Content without a point of view is just a summary, and, as he puts it, nobody wants to read an AI retelling of a podcast they could have listened to themselves.
Once he has his take, the rest is fast.
For Twitter, he tells Verdent what he thinks and runs:
/write twitterVerdent distills the transcript highlights plus his POV into an 8–15 tweet thread. First tweet gets a hook, last tweet gets a CTA, every tweet stays under 130 characters, and direct quotes from the transcript come with timestamps. Output drops into content/YYYY-MM-{slug}/platform/twitter.md.
For LinkedIn:
/write linkedinAn 800–1500 word article — hook up top, two or three core insights with his commentary in the middle, a closing question to drive engagement. The tone is professional but personal. Not a press release, not a tweet. What Ethan has learned about LinkedIn specifically is that podcast takeaways framed through personal professional context — "5 years doing X, first time I've heard someone say Y" — outperform generic shares by a wide margin.
For Instagram:
/write instagramTen to fifteen carousel pages: title, key insights, pull-quotes, CTA. Verdent generates the HTML and exports 540x720 @2x PNGs into exports/, ready to post. Copy and hashtags land in platform/instagram.md.
Then the long version — which is usually where Ethan actually wants people to end up:
/write blogA 1500–3000 word post with subheadings, inline transcript quotes (with timestamps), and his analysis. The blog is the most complete version — Twitter and LinkedIn drive traffic to it. The post auto-links back to the original podcast and credits the speaker.
Last thing Ethan does is one quick pass across everything: check the quotes are accurate, his point of view is clear, no leftover AI-style filler. If it reads well, he ships it.
What Changed
| Before | After | |
|---|---|---|
| Audio → text | Listen at 2x (still 45 min for a 90-min episode) | /transcribe → transcript in ~2 min |
| Finding highlights | Notes in a separate app, context lost on every app switch | Skim + mark 3–5 moments, 5 min |
| Writing for platforms | Twitter ~20 min + LinkedIn ~30 min + blog ~60 min, each rebuilt from scratch | /write → 4 platform-native versions in ~10 min |
| Total | 3–4 hours, often abandoned halfway | Under 20 minutes, all inside Verdent |
Things he figured out the hard way
A few things that took him a couple of cycles to get right:
- The skill is yours to evolve. Since /transcribe is a skill Ethan wrote himself, whenever he hits a gap — a new platform, a missing metadata field — he just tells Verdent to add it. The workflow grows with him instead of waiting on a product roadmap.
- **ASR errors make the cleanup prompt deceptively detailed. **Speech recognition will confidently turn "gRPC" into "GRPC," "Envoy" into "envoy," and speaker names into near-misses. The cleanup prompt catches most of it, but Ethan still does a quick scan of the cleaned transcript before running /write — a couple of wrong proper nouns can poison every downstream output.
- **You don't have to cover the whole episode. **A lot of two-hour interviews have one really good 30-minute stretch. Ethan will tell Verdent: "focus on the 45:00 to 1:15:00 segment about X." Output is way tighter, and the post lands better.
- Multiple episodes on the same theme are great fuel. Transcribe each separately, then ask /write to synthesize across sources — "What I learned about distributed systems from 3 podcast episodes" type pieces tend to do really well.
