Engineering Management: The Bottleneck Is the Organization

The host introduces the guest as a man who has spoken at nearly every Russian conference; Anatoly corrects him: mobile ones are still ahead. The conversation opens with childhood: born and raised in Moscow to two engineer parents, he had to wait another two or three years for the promised computer because of the 1998 crisis. He wanted to make computer games — the usual motivation — yet never worked in the industry, and built his first one only this summer, with his son, in Cursor. He went to study programming at the steel and alloys university, where the only programming on offer was Basic in the early years, and his diploma turned out to be a business plan for a pipe-rolling plant. By the second or third year it was clear he was being taught the wrong thing, so he took evening work at a web studio building sites out of tables; today's rows and columns, he notes, look suspiciously similar. The studio closed, he taught himself PHP from a book — there were no courses then — and joined a company selling ringtones and pictures through short numbers.

His favourite project of those years was a small social network of his own, with a forum and a chat, popular with mobile phone owners: to see a new message you refreshed the page by hand. Alexander recalls his own company building two social networks back then: the one assembling its feed by joining a relational database collapsed as soon as real users arrived. Then came a property classified, among the first to shoot 3D panoramas of courtyards, and a hosting business with admin panels for cloud providers — a year to eighteen months in each, as he recalls. Recommendations from friends brought Anatoly to Badoo, where the interview, by his own reckoning, went fifty-fifty. He spent five years in a single team — billing and payment provider integrations — and grew from junior to senior with no grading system: career ladders and performance review came with a CTO who arrived from Google late in that period.

For his last couple of years Anatoly was a tech lead: his manager owned projects and people, he owned engineering; one of the big projects he recalls was a separate infrastructure for accepting payment cards under PCI DSS, which he ran as a senior engineer: code from scratch, a secure perimeter built with the sysadmins, and homemade bash delivery that scattered an archive across machines; preparing for the interview, Alexander found a recorded talk about it. He left without conflict: he wanted to grow as a manager, and the leadership positions around him were all taken. Lazada appeared just then: an Asian e-commerce business that, on foreign funding, counted salaries in dollars, which after yet another currency jump looked especially attractive; this was around 2014, by the guest's recollection. He hesitated a long time; his wife tipped the balance: however bad it got, they would live through it together. Alexander unpacks that fork — a brain used to counting options builds a decision tree with expected values and concludes the game is not worth the candle.

The Moscow hub had been opened for a specific pain: the company had a Russian CTO and VP of engineering, the Vietnamese hub had run out of strong engineers, and seasonal sales produced a peak load the PHP application could not take. When he interviewed, the hub held about twenty people against a plan of a hundred by year end; in nine months the team grew fivefold. Everything happened at once: hiring, processes, and building the Go infrastructure from nothing. Go itself proved both a hiring magnet — people came tired of PHP, C# and .NET — and a problem: they wrote the new language with the habits of the old, which the team recognised about a year in and treated with scheduled refactoring. Docker and Kubernetes arrived in the same period, picked up straight off their releases, while telemetry was written in-house because Prometheus did not exist. Strict processes came last: strong engineers kept arriving from the market, and the teams largely organised themselves.

Both men compare rebuilding on the move to repairing a plane in mid-air. Later Alibaba bought Lazada, the Moscow hub closed, development moved to Asia, and Anatoly left for Avito ahead of the rest, giving up the project bonus; a couple of months later Ozon offered his former colleagues a move of their own. Avito invited him into another attempt at growing verticals — niche segments such as property, cars, services and jobs — on the second-player logic of biting segments off a universal leader one at a time. What he inherited was modest: a rapid-response team, the Avtoteka team with its VIN-based car history, and Domofond, a property classified sitting third or fourth nationally. Unlike Lazada, the focus was organizational: grow the number of teams and work out how verticals mesh with horizontal teams improving search and filters for everyone.

Alexander brings up Conway's law; Anatoly replies that in the early years the monolith saved them — everything needed could be fixed in one repository — and that it got harder once microservices arrived. The first pebble of change was the info model — a database holding the tree of categories and ad attributes: without engine displacement or floor area no vertical experience is possible, yet adding an attribute took a year when ten were needed. The flexible schema came from the data warehouse team, and Anatoly led its integration into the key services. Alongside it, a catalogue of a hundred thousand car variations simply refused to let the PHP process start. At first change was sold by voice and slide decks, with ADRs and one-pagers coming later; one of the key levers, as he sees it, was the verticals' top managers, with whom he presented a united front. Alexander adds his own case: a web account area the business saw no point in funding, where architecture work done in advance proved very useful once the context changed. Anatoly replies that partnership trust with the business and product leader is what lets you make such calls autonomously and invest before anyone asks.