Kokotov, Dan
▶ Watch on YouTubeCore ideas
Kokotov explains Rev's origin as a deliberate improvement on the Upwork freelance-marketplace model: rather than forcing non-expert customers to evaluate freelancer profiles (a high-friction, regret-prone process for both sides), Rev's founder — an early Upwork employee — chose narrow, standardizable verticals (starting with document translation, later audio transcription and captioning) where customers simply submit files and receive finished output without visibility into the underlying process. Kokotov frames Rev as a two-sided marketplace whose central operational challenge is balancing supply and demand between customers and "Revvers" (freelance transcribers/translators) — too few Revvers slows customer turnaround, too many leaves Revvers without enough available work — and pushes back on the term "gig economy," which he feels trivializes the work, preferring to frame Rev as enabling flexible, location-independent work for a demographic ranging from stay-at-home parents to socially anxious individuals to people deliberately taking career breaks.
He describes Rev's actual transcription workflow: Revvers select audio jobs matching their preferences (length, subject, dialect) in a specialized editor, starting from an automatic speech recognition (ASR) draft that they correct rather than transcribe from scratch — except for poor-quality audio (e.g., far-field lecture recordings with crosstalk), where working from scratch may be faster. He estimates realistic human transcription speed at roughly 2-3x real-time even for good audio, correcting Lex's hope for near-real-time turnaround. On the technical core, Kokotov defines ASR as the machine learning task of converting speech audio into text, contrasts general "unstructured" transcription (Rev's focus) with narrow-vocabulary voice-assistant ASR (e.g., Siri), and states Rev's current word error rate is around 14% against human performance of roughly 2-3%, calculated via insertions, deletions, and substitutions relative to a reference transcript. He identifies proprietary training data — specifically the fact that Rev's own paid human-correction workflow generates enormous volumes of high-quality labeled audio-transcript pairs — as the company's central competitive advantage (a "flywheel" he says is still only partially exploited, since Rev also retains detailed edit histories that could in principle provide additional training signal beyond final transcripts alone), and positions Rev.ai as the company's platform play: exposing its ASR engine as an API for third-party developers to build on, modeled loosely on AWS's approach of releasing infrastructure primitives and learning from what people build with them.
The conversation moves into a broader critique of large tech companies' product and community stewardship: Kokotov and Lex jointly criticize Amazon Mechanical Turk's neglected interface and outdated API documentation, and extend this to a critique of YouTube's perceived unresponsiveness to creator requests (citing the long-missing multi-video-upload feature) as a symptom of large organizations optimizing for aggregate metrics (e.g., total watch time) rather than genuine care for the people the product serves — contrasted with Steve Jobs's obsessive, detail-level personal review process at Apple (referencing Ken Kocienda's book Creative Selection) and Elon Musk's similarly hands-on engagement with low-level engineers at Tesla, which Kokotov frames as a rare, hard-to-replicate leadership style requiring enough independent credibility to defy short-term investor pressure.
Kokotov and Lex discuss the broader vision of universally transcribed and searchable audio — meetings, podcasts — as a potentially transformative shift enabled by accurate ASR, lamenting that podcast discovery and reference currently lag far behind text-based media (comparing it to early pre-search-and-discovery YouTube), and discussing an emerging RSS-based standard (raised by a company called Buzzsprout) for embedding transcripts directly in podcast feeds, which Kokotov and Lex both frame as preferable to platform-exclusive models like Spotify's Joe Rogan deal, with Lex expressing a preference for open RSS distribution ("the pirate radio aspect of podcasting") even at the cost of forgoing exclusivity payouts. This leads into a discussion of platform moderation and free speech, where Kokotov articulates a preference for platforms encouraging kindness and healthier discourse ("carrot" incentives) over censorship, alongside skepticism that removing content meaningfully changes deeply committed believers' minds, while acknowledging the difficulty platforms face in distinguishing dangerous conspiracy theories from more benign ones, and noting that engagement-driven business models structurally reward outrage over long-term user well-being — a dynamic he believes could be reversed by optimizing for long-term user happiness rather than short-term engagement, though this requires rare founder-level independence from short-term market pressure.
On his own career transition from individual-contributor programmer to engineering manager, Kokotov describes losing the direct, tangible "dopamine" feedback loop of writing code and having to learn to derive satisfaction vicariously through team success, and reflects on tailoring management style to individual psychology — some people responding to blunt criticism, others needing consistent encouragement because they are already highly self-critical — citing the management book First, Break All the Rules and its "manage by exception" philosophy as influential at Rev. The conversation closes with book and film recommendations (all six main Dune novels, praised as a single continuous chronological story; Huxley's Brave New World, discussed as prescient about genetic and ideological "sorting" into echo chambers; Orwell's 1984; and the film Brazil, which Kokotov considers a more realistic dystopian vision than 1984 because it depicts authoritarianism as bureaucratic incompetence rather than efficient evil, a theme he connects to HBO's Chernobyl), reflections on podcasting as a strange, evolutionarily-resonant one-directional (and sometimes eventually mutual) human bond, and a closing exchange on meaning — Kokotov, describing himself as not religious, locates meaning in contributing to humanity as a kind of collective evolving organism through creation (explicitly including small acts like programming) and raising children who pursue their own unique interests.
Excerpts
On Rev's founding insight versus Upwork:
“Rev was kind of founded to improve on the model of Upwork... it was kind of a difficult experience because you would have to browse through all these people and decide, is this guy good or is somebody else better?... Let's remove the barrier — we'll pick a few verticals that are fairly standardizable... and we'll just make it a website: you give us your files, we'll give you back the results as soon as possible.”
On why proprietary correction data is Rev's core advantage:
“We believe in ASR the biggest thing is the data... the more data you have and the higher quality the data, the better labeled the data... we at Rev have kind of the best data — like, your literal business model is being paid to annotate the data, so it's kind of like a pretty magical flywheel.”
On why Brazil feels like a more plausible authoritarian future than 1984:
“It's a dystopian movie of authoritarian incompetence — nothing really works very well, the system is creaky, but no one is willing to challenge it, things kind of amble along. It's not this super-efficient evil dictatorship of 1984, it's just kind of this badly functioning... status quo that just goes on.”