Video summary

Школа AI агентов на локальных моделях, день #3: Разбор ТОП-20 AI-агентов и рынка вакансий

Main summary

Key takeaways

Educational

Main ideas / lessons

1) Why AI agents matter (and how to think about building them)

  • The speaker positions “AI agents” as practical units you can implement to automate work, sell to businesses, or create your own services.
  • The key skill for generating useful agent ideas is observation:
    • Watch many existing agents (the “neural network” effect: your brain absorbs patterns).
    • After seeing enough examples, ideas emerge naturally for new projects.
  • The webinar focuses on employment, freelancing, and monetization, not just building agents.

2) Market framing: where demand comes from

  • Agents are categorized by who pays/uses them:
    • B2C (sold to individuals)
    • B2B front office (interacts with customers externally)
    • B2B back office (works inside the company)
  • The speaker emphasizes that profitability depends on mapping agent capabilities to real organizational needs.

3) A structured set of agent ideas (the “TOP list”)

  • The speaker claims they prepared 70 agent examples organized systematically (from potentially hundreds).
  • Examples are grouped into domains, including:
    • Education: language tutors, programming mentors, homework checking, course creation
    • Health & fitness: trainers, nutritionists, advice, analysis
    • Psychology & self-development: coaches, time management, consultants
    • Corporate internal consulting: burnout/psychotyping, role fit, internal evaluation
    • Shopping/selection assistants: equipment, gifts, stylists
    • Finance: budget planning, expense control, investment assistants
    • Travel: route planning, city guides
    • Entertainment & content generation: children’s fairy tales, music generation, recommendations
    • Home & family / pets: design, parenting help, veterinary-style guidance

4) “Do it for yourself” vs “sell to order”

  • For B2C, you can start by building for personal use.
  • For selling to companies, it becomes B2B front office: the company needs measurable improvements (retention, service quality, cost reduction).
  • Warning about marketing reality:
    • B2C subscriptions are hard to scale without heavy budget, PR, and branding.

5) Core monetization paths (four directions)

The speaker explicitly lays out multiple “tracks” and how they relate to money:

  1. Projects for yourself (portfolio + internal capability building)
  2. Career growth inside a current profession (non-obvious track)
  3. Employment (stable job)
  4. Freelancing / selling services (projects or subscription services)

Detailed instruction-style methodology and frameworks

A) The “observation → idea generation → practice” loop

  • Observe many agent examples:
    • “Soak it up” from lots of agents to form patterns.
  • Practice by building variants:
    • Start with many ideas but implement some to gain a high-level understanding.
  • Re-test your observations:
    • When you build with your own hands, you “confirm” what you learned and improve idea quality.

B) How to choose a monetizable agent/service topic (especially for B2B subscriptions)

Practical selection rule:

  • Pick a topic that creates a specific business benefit:
    • Preferably expressible in rubles, or tied to a concrete pain point.
  • Validate with real potential buyers:
    • Talk to 10 potential buyers about the concept.
    • Expected signals:
      • Many should respond: “I would buy this for ~10,000/month”
      • 1–2 may already ask for timing and implementation details
  • Avoid vague ideas (common failure pattern):
    • Don’t propose generic “a service for lawyers” without:
      • What exact lawyer work it reduces
      • Load reduction % and why that’s worth paying
      • Who the buyer is and where the budget comes from

C) Freelancing: project-order funnel (sales process)

A typical pipeline for custom AI project sales:

  • Funnel stages (approximate, with drop-offs):
    • 20 initial contacts
    • 100 emails sent (example-based approach)
    • 2 responses → Zoom discussions
    • 10 Zoom consultations
    • 5 evaluation-stage drop-offs
    • 3 disappear due to price/shift/objections
    • 2 proceed to contract + technical specs coordination

Sales is framed as:

  • Expert-driven discovery and consulting, not performative “CEO meetings.”

The speaker’s typical workflow:

  • Ask lots of questions
  • Provide small technical consulting
  • Produce a project estimate (e.g., after 2–3 days)
  • Negotiate on TZ/spec, price, and payment terms

D) Freelancing: project-order vs subscription (“hunter vs farmer”)

  • Project order (Hunter):
    • Catch a “wild boar” once
    • Income comes in bursts; you must keep “hunting”
  • Subscription service (Farmer):
    • “Plant” once; harvest slowly
    • More stable recurring income if you find clients and retain them

Employment-focused key points

A) Hiring reality: jobs exist because of AI adoption gaps

  • The speaker claims the main barrier for companies implementing AI is finding/attracting specialists with the right competencies.
  • Labor market claims mentioned:
    • AI hiring growth, remote work demand
    • Shortage of developers/AI talent

B) What helps get hired (3 recommendations)

  1. Be active:
    • Interviews, tests, consistent follow-up—don’t disappear
  2. Be socially adequate:
    • Punctual, appropriate behavior, avoid chaotic backgrounds/noise
    • Avoid “evasive answers” (framed as HR red flags)
  3. Build a portfolio with real projects:
    • Portfolio project types:
      • For yourself (practice/learning)
      • For a company (implemented)
      • Through internships (real company project + official letter of gratitude)

C) Internships and “real project proof”

Internships are described as:

  • 3 months
  • Team size: 15–20 people
  • Delivered as real-world projects, accompanied by gratitude letters

Examples named include projects for large Russian organizations (e.g., telecom/rail/industry/academia/auto-related contexts).

Tech/implementation perspective (local models + platforms)

  • The speaker contrasts cloud vs local:
    • Companies with strict security may block cloud usage.
    • Local models deployed on Russian infrastructure can be exposed via tokens.
  • Tools mentioned:
    • Wipecoding / cloudcode (code generation/development platforms)
    • Replit and cloud-style deployment
    • Mentions of running speech/image recognition pipelines as demos.

Agent development maturity model (3 levels)

As agent-building becomes more real, the speaker describes three roles:

  1. Razrap/Prompt (early stage: prompt-level development)
  2. Archmage (architect): designing structure/system integration
  3. Integrator: embedding into business processes

Forecast:

  • The market shifts from “build an agent” to integrate it properly into real workflows.

Webinar logistics / promotional mechanics (draws, bonuses, access)

  • There were two giveaways:
    1. For everyone who registers
    2. For people who submit at least two homeworks
  • Registration collected name/email/phone.
  • A “roulette” mechanism is used to distribute bonuses/courses:
    • Spinning gives discounts/courses/internship-related perks
    • The speaker claims non-guaranteed acceptance rates for some buyout/credit-like offers, with typical statistics (e.g., “~22% accepted” for roulette outcomes).

Main speakers/sources featured

  • Single primary speaker/host: the lecturer leading the webinar (name not given in subtitles).
  • No other clearly identified speakers: guests/other mentors are mentioned generally, but not individually named as speakers.
  • Referenced external sources/organizations (not as speakers):
    • HabrCareer (salary data source)
    • Yandex and partners (research on AI adoption barriers)
    • Ministry of Digital Development (developer shortage claim)
    • Universities/partner companies cited: Synergy, MTS, Sberbank, Tinkoff (and others implied/listed in passing)
    • Platforms/tools mentioned: Telegram, VK, WhatsApp, Yandex Maps, HeadHunter, plus tools in the style of Google Studios / Replit / Kloudcode as services

Original video