Pavel Durov on Building Telegram, Privacy, and Platform Independence

Guest:
Pavel Durov — Founder and CEO, Telegram
Host:
Lex Fridman
Source:
Lex Fridman Podcast · 30 September 2025

Pavel Durov on Building Telegram, Privacy, and Platform Independence

Pavel Durov, founder of VK and Telegram, spends nearly four hours and forty minutes with Lex Fridman on personal discipline, the engineering philosophy behind running a billion-user platform with a 40-person team, Telegram’s privacy architecture, and his account of state pressure on the company culminating in his 2024 arrest in France.

Key ideas

  1. Deliberate understaffing forces automation. Telegram’s core engineering team is roughly 40 people managing about 100,000 servers. Durov holds the headcount deliberately low — past the point the company could easily afford more — reasoning that a larger team spends its time coordinating rather than building, and that the constraint forces the automation that makes the system scalable in the first place. See Deliberate Understaffing.
  2. Privacy by architecture, not by policy. Since 2012–13, Telegram has been built so that no employee, Durov included, can access users’ private messages by design. Opt-in end-to-end encryption (‘Secret Chats’, launched 2013) and reproducible open-source builds on both Android and iOS let the claim be independently checked rather than merely asserted.
  3. Ownership as the enabling structure. Durov owns 100% of Telegram; there are no outside shareholders. He presents this directly as what lets the company hold a costly privacy commitment, and refuse government data requests, without a shareholder base able to override the decision.
  4. Personal discipline as an operating philosophy. No alcohol, no habitual phone use, daily physical training, and a stated preference for self-restriction over abundance recur throughout as the same principle applied to the self that Telegram applies to its engineering: remove what does not serve the goal.
  5. State pressure and the 2024 arrest in France. Durov gives his own account of an August 2024 arrest and ongoing French investigation, alongside earlier pressure from Russia, Iran, and Apple. This is recorded here as his contested, one-sided account of an active legal matter, not as settled fact.

Content

Building lean: Telegram’s engineering philosophy

Durov’s account of running Telegram rests on a claim most fast-growing technology companies reject in practice: that adding people to an engineering team past a certain point makes the product worse, not better. His reasoning has two parts. Coordination cost rises with headcount, so a larger team spends more of its time synchronising and less time building. And a hard headcount ceiling is itself a forcing function — refusing to hire compels the existing team to automate, which is what actually lets roughly 40 engineers run infrastructure for over a billion users across some 100,000 servers. He adds a morale argument to the efficiency one: an employee without enough real work to do tends to manufacture problems rather than sit idle, and he describes several cases where removing a single underperforming engineer raised a team’s output, because the cost of working alongside weak performance outweighed the raw capacity lost.

Hiring runs through public coding competitions rather than conventional recruiting — a mechanism Durov says filters for ability under a genuine constraint and screens out candidates used to a large company’s diffuse accountability. VK itself was written by Durov alone for its first year, starting, on his brother’s advice, with nothing but a login module: get one small thing fully working before building anything else. The same minimalism shows up in what Telegram removes rather than adds — ‘getting rid of features is super important,’ in his words — a claim about product focus that echoes Do Half‘s feature-cutting principle in a different domain: engineering load and onboarding simplicity rather than usage-metric concentration. See Deliberate Understaffing for the fuller mechanism.

Privacy and encryption architecture

Telegram launched opt-in end-to-end encryption (‘Secret Chats’) in 2013, which Durov dates as roughly fourteen months ahead of WhatsApp’s equivalent. The encryption is opt-in rather than universal because, in his account, full end-to-end encryption cannot support the features Telegram’s scale requires: persistent multi-device history sync, very large group chats, and bot integrations. Ordinary (‘cloud’) chats remain encrypted but are, in principle, decryptable by Telegram — a trade-off he frames as a deliberate choice between usability and maximal confidentiality, made explicit to users rather than hidden.

The architecture’s more distinctive claim is that no Telegram employee — including Durov — can access private message content by design, a property the company says it has held since before it added end-to-end encryption at all. To make that claim checkable rather than simply asserted, Telegram has published its app source code since 2013 and, per Durov, is the only major messaging app offering reproducible builds on both Android and iOS: independent verification that the binary a user downloads matches the code publicly visible on GitHub. Much of the underlying stack — database engines, web servers, even the back-end API language — is built in-house rather than assembled from third-party open-source components, which Durov ties to minimising the number of external dependencies that could carry an undisclosed flaw.

Platform independence and censorship resistance

Durov frames Telegram’s ownership structure as the mechanism, not merely the outcome, of its privacy stance: he owns the company outright, took a salary of roughly one dollar, and says Telegram has never returned him more in personal income than he has invested in it. Without outside shareholders, he argues, there is no board able to override a refusal to hand over user data or to comply with a takedown request Telegram considers a matter of political speech rather than its actual content rules (which do bar incitement to violence and material such as terrorism and child-exploitation content — areas where he says Telegram runs an active, machine-learning-assisted moderation programme with public transparency reporting).

He narrates several specific episodes of government pressure: Russia’s 2018 blocking attempt, evaded through a crowd-sourced proxy network he calls ‘Digital Resistance’; a month-long dispute in which Apple withheld Telegram’s App Store update pending resolution with Russian authorities, resolved only after public pressure; Iran’s 2018 ban, worked around through an economic incentive scheme for citizen-run proxy servers; and — the most serious and most contested episode in the conversation — his August 2024 arrest in France and subsequent investigation, during which he alleges French intelligence separately asked him, through an intermediary, to restrict political channels ahead of elections in Romania and Moldova. He says he refused requests concerning peaceful political speech while complying with requests that matched Telegram’s own content rules. This wiki entry records the arrest and pressure allegations as Durov’s own, one-sided account of an active and legally contested matter — no French official or counter-account appears in the conversation — rather than as adjudicated fact. Fridman closes the published transcript with his own appended essay drawing a parallel to Franz Kafka’s The Trial, a framing device added after the interview itself.

Personal discipline and creativity

Durov ties his personal regimen — no alcohol, minimal pharmaceutical use, deliberate avoidance of a smartphone for ordinary use, daily physical training including 300 push-ups and squats, regular cold exposure and long open-water swims — to the same underlying principle as his engineering philosophy: identify what does not serve the goal and remove it, rather than treating its symptoms. He frames avoiding a phone specifically as protecting his own attention from being set by other people’s algorithms, which he presents as directly instrumental to running Telegram rather than incidental self-care. On abundance, he draws on Universe 25 (the 1960s–70s ‘mouse utopia’ experiments) to argue that self-imposed restriction, not the removal of all constraint, is what produces purpose and sustained output — a frame he explicitly extends to AI-driven abundance as a live risk for humans generally.

See also

See also