Video summary

Правда об ИИ: кого уволят первым? Дебаты бэкендера, MLера и аналитика

Main summary

Key takeaways

News and Commentary

Main thesis: AI automates tasks, but roles shift rather than disappear

The panel discusses how AI is expected to reshape the labor market for:

  • Backend developers
  • Analysts
  • ML / Data Science specialists

Core assumption: AI automation will grow over the next 5–10 years, but it won’t necessarily eliminate all jobs. Instead, it will change which parts of each role remain valuable.

A recurring idea is that “copypaste”/routine parts are the most automatable. Work that requires real judgment, accountability, and domain/business understanding is harder for AI to take over completely.

Claim 1 (Pavel / Backend): Analysts are the first to go; backend is last

Pavel argues that backend engineers will be replaced later than analysts, claiming:

  • Many analysts’ work is already “artificially intelligent” in practice—outputs often look like AI-generated reasoning.
  • Backend is closer to real engineering constraints and complexity.

He also criticizes the other panelists’ arguments and suggests that many people in analysis roles are closer to “students” whose work can be outperformed by AI in at least some portions.

Claim 2 (Nikita / Analytics): Analysts stay afloat longer due to human judgment and “impression”

Nikita counters that analytics is likely to remain viable, especially because:

  • Analytics involves human psychology, interpretation, and communication with business stakeholders.
  • Even if AI can generate answers, it may struggle with the nuance of what truly fits the business context.
  • Analytics remains important where AI-generated outputs must still result in convincing decision-making.

He also emphasizes that while AI may reduce some routine tasks, analysts can remain central when decision-making and methodology coverage matter.

Claim 3 (Makar / ML): ML specialists will last longest; AI can’t replace core ML thinking end-to-end

Makar argues that ML / Data Science roles will be more resilient than backend or analytics because:

  • ML requires human decisions: model design, hyperparameters, reasoning behind choices, and production deployment.
  • AI can’t reliably “derive the model” or guarantee correct behavior without human accountability.

He frames ML/DS engineers as “teachers”/leaders who guide AI-enabled work within a broader hierarchy.

Discussion of automation by sector (and critique of the evidence)

A slide/graph claims roughly ~90% automation pressure in areas like:

  • business/finance
  • computer science/math
  • management

…and less automation for:

  • food/service
  • agriculture

Critiques raised

  • Makar: The graph is poorly constructed and mixes categories unfairly, such as:
    • conflating different mathematical/computing domains
    • ignoring robotics/labor tech (e.g., agricultural drones, automated greenhouses)
  • He argues this is likely bait for simplistic fears, because AI labor-tech differs from “AI research.”
  • Nikita: Challenges the methodology and highlights coverage/task scope: even if AI can automate parts of a domain, the real question is how much of the job’s task coverage gets replaced.

Technical reasons it won’t fully replace: security, tooling, and bureaucracy

The panel points to obstacles that make deployment less seamless:

  • In large companies, AI may be restricted by data security and internal policies
    • e.g., inability to use external GPT services
    • closed/isolated corporate knowledge
  • Even if “agents/MCP servers” exist (mentioned via a Yandex-like architecture), integration depends on:
    • internal systems
    • documentation accessibility

Makar and Pavel also note that custom internal code/libraries can limit AI usefulness if the model can’t access the right documentation.

“Turing test”-style claim: AI can do only a small fraction of a worker’s tasks

A study is mentioned where a model handling a simulated working day completed about 6–7% of tasks (described as “intern” performance). This is used to support the idea that even strong code-writing AI can’t yet fully run end-to-end professional workflows.

Hierarchy and “who replaces whom”

The panel debates whether backend or analytics is closer to elimination.

Key synthesis

  • Rather than pure replacement, roles may converge into a combined AI-enabled specialist (ML engineer + analytics + backend tasks).
  • Some routine analysis/implementation steps could be absorbed into unified workflows (a new “all-in-one” stack).
  • However, accountability and responsibility still require human oversight.

There is also disagreement about which discipline has the most replaceable “top of the funnel” tasks (juniors/intern-level work).

New roles that may appear

The video claims additional AI-related professions may grow, including:

  • MLMOps / ML engineer roles focused on deployment and integration
  • AI transformation leaders driving AI adoption in organizations
  • reviewer / AI-assisted code review and automation of routine development checks
  • expanded roles around operating LLM-based systems in production (frameworks, pipelines, governance)

What will actually be automated in analytics and backend

Analytics

AI is expected to automate:

  • generating SQL / SKEL-like queries
  • parts of statistical analysis setup

The “hard part” remains:

  • methodology choice
  • defining tests
  • connecting models to real-world/physical/business meaning
  • building features that capture subtle behaviors
    • example: micro-trends like short-form video viewing affecting ad strategies

Backend

AI may accelerate:

  • routine coding
  • glue code
  • “spec-to-code” tasks

But complex architectural constraints still require domain experts.

Career advice: juniors will face the biggest squeeze; focus on “real value”

All speakers argue junior roles are most at risk because:

  • AI reduces entry-level routine work (query writing, basic implementation tasks, etc.)
  • fewer junior openings may remain as companies demand faster, more senior outcomes

Recommendations

  • Gain experience quickly (internships, verifiable projects)
  • Build practical credibility (portfolio, real work, strong interviews)
  • Move toward deeper, complex domains
    • example: backend complexity like databases / distributed systems

Interview cheating and AI-assisted hiring concerns

Makar describes AI-enabled cheating during interviews, including:

  • candidates using external prompts or screen-reading tools to generate answers live

He notes detection is difficult if the process isn’t closely monitored. The panel still argues that long-term success requires true competence; interview questions about real past work may expose bluffing.

Closing: promotional insert about courses

The discussion ends with a strong promotion for the speakers’ courses (algorithms, backend/ML/analytics), including a stated discount and promo code. The speakers deliver final wishes, encourage subscribing to channels, and invite contacting managers.

Presenters / contributors

  • Pavel Ivanov — backend developer; course author/host
  • Nikita — analytics teacher; course leader
  • Makar Artyomov — ML / data science specialist; course author
  • Alexander — mentioned as a Telegram manager handling promo/QR code

Original video