How AI killed outsourcing
Как AI убил аутсорс
B · Useful, limited Podlodka podcast with the owners of a Russian mobile studio and the US-focused consultancy Evil Martians giving first-hand market signals; concrete and candid, but anecdotal and partly based on the guests' own unverified readings.
Summary based on the full transcript·Original: about 116 minutes to listen
In short
Podlodka Podcast episode 497 (5 October 2026, Russian, 116 minutes). Two owners of development-services firms describe their markets after AI coding tools. Denis Germanenko (CleverPumpkin, Russian mobile studio since 2011) covers Russia: shelved tenders, collapsing rates, no junior hiring. Irina Nazarova (Evil Martians, US startups) covers the US: teams shrinking from 5-7 people to 1-2, founders shipping huge AI-written codebases, and buyers asking for headless products their own agents can use. Both reject the large productivity claims.
Key numbers
- ~5%profitability of Russian custom development in 2025, per the guest's reading of public filings, against 15-20% seen as healthy
- tens of timesfall in inbound leads reported by the Russian studio
- 10 to 2-3people now needed on a project that took a team of ten five years ago (Evil Martians)
- at most 2xIrina Nazarova's estimate of the real individual productivity gain from AI; 10x claims called a bubble
- 50-70%margin the Russian guest thinks possible when selling outcomes instead of hours
As reported by the source.
Russian·Video·opens youtube.com
What it reports
- Russia: "phantom tenders" run to the end then cancelled for lack of budget; clients expect much lower prices because "AI makes development cheaper" and question hour-based estimates.
- Requests shift from "build an app" to "raise revenue or cut cost", with payment tied to outcomes.
- Inbound leads down "tens of times"; custom-development profitability about 5% in 2025 while costs (VAT, reduced IT tax benefits, salaries) rose.
- CleverPumpkin shrank production to small teams for two or three long-standing clients; resistance to AI came most from the most experienced staff.
- Studios no longer hire juniors because agents do their tasks; Germanenko expects a shortage of solid mid-level engineers in 2-3 years. "Phantom vacancies" keep recruiters busy.
- Neither guest has seen a studio win many clients through AI productivity; fifteen parallel agent sessions for a week burnt tokens without shipping features.
- US: founders who never coded report 80 PRs a day and arrive with projects of 100k to 1M lines "90% done". Evil Martians separates disposable code and gives clients harnesses (design system, linters, hooks) so their agents stop degrading it.
- A project that needed ten people five years ago now needs two or three, but Nazarova puts the individual gain at "at most 2x".
How they got it
Interview with two practitioners. Transcript from cleaned YouTube auto-captions, with some names garbled. The 5% margin is the guest's figure from public company data, not shown on air.
Limits
- Two firms, two markets, opinions and anecdotes; no data shown on air.
- The Russian picture mixes AI effects with macroeconomic and tax changes, which the guests themselves cite.
- Recorded before the grading window (ext item); auto-caption errors are possible in quotes.
Why it matters for companies
For buyers of outsourced development, the bargaining power has shifted: suppliers are open to outcome-based pricing and smaller embedded teams. The warnings are useful too: AI-generated codebases arrive large and fragile, and the fix is harnesses and guardrails for the client's agents, not rewrites. The end of junior hiring in services firms will feed a mid-level shortage that every employer will feel in a few years. A grounded estimate of about 2x per person is a better planning number than vendor claims of 10x.
Key passages in translation
Russian original
маржинальность компании на заказной разработке около 5%. То есть рентабельность 5%.
The margin of companies in custom development is about 5%. That is, profitability of 5%.
Russian original
Мы не можем ничему научить самих людей, но мы можем научить их агентов чему-то.
We cannot teach the people themselves anything, but we can teach their agents something.
Russian original
Не знаю, максимум в два. Я я бы сказала.
I don't know, at most two times. I'd say.
Summary based on the full transcript. Read the original for the full argument and data.Watch the original on youtube.com