18.09 Algorütm: Vajalikud arendajate oskused, millest keegi ei räägi

Core framing: technology won’t remove engineering friction

Even with advanced technology, software systems and components often don’t integrate cleanly. Teams still need hands-on debugging—reinforcing a “Murphy’s Law” reality for engineering work.

What the episode is about (digital stage / 2026 skills theme)

The discussion centers on which developer skills and competencies matter in the software industry—especially in an AI-era—including several that are rarely discussed publicly.

Where the curriculum direction comes from (market research behind “ee code” / Kood)

The speakers describe curriculum updates as a product-cycle approach:

• The “product” isn’t only the learning platform.
• The product outcome is the graduate entering the job market.

Market research inputs

They cite:

• Sample size: 185 employers have hired graduates from Code/ee-code.
• Breadth: hiring spans traditional software houses and AI-native companies, plus product companies and even sectors without in-house software developers.
• Role coverage: feedback came from recruiters, senior engineers/managers, and alumni.

The goal is to identify:

• common pain points,
• hiring indicators,
• what the curriculum should solve.

Misleading job ads vs real hiring criteria

Job ads may list lots of tech keyword checkboxes, but the research suggests hiring is less about matching a checklist.

A consistently cited deciding factor for junior candidates:

• Motivation—described as a “sparkle in the eyes” and a willingness to develop.

While tech fundamentals remain crucial, they are not sufficient to differentiate candidates.

How to teach “basic AI knowledge” without it becoming obsolete

The recommended approach is to teach:

• stable principles and
• reasoning skills

rather than transient details about specific tools or agent implementations.

They specifically avoid overemphasizing:

• quickly-changing “coding agent” specifics.

Instead, they recommend:

• teaching through questions and experimentation,
• covering durable LLM fundamentals such as next-token prediction and conceptual differences, rather than memorizing product/tool output.

AI engineering verification: tests, evidence, and evaluation change

They argue that agent/LLM interactions can be hard to assess with traditional intuition like compute complexity. Often, you need experimentation to discover failure modes.

Implications for engineering practice:

• shift from relying primarily on traditional “unit/integration tests”
• toward evidence/validation for AI/agent workflows
• emphasizing outcomes, robustness, and whether the system is trustworthy

Measuring and nurturing candidate qualities (“sparkle” persistence)

To operationalize motivation, they added a motivation validation point to intake tests.

Since persistence is hard/impossible to teach directly, they try to:

• screen for persistence
• also assess traits during intensive intake periods

Junior advantage in the AI era

Juniors may sometimes outperform seniors because of:

• openness to experimenting with new workflows and tools
• ownership and “readiness to learn” (seniors may feel too comfortable)

Seniors still matter for:

• code review quality,
• “cleaning work,” and
• knowing what not to let through.

Communication and “product thinking” as major differentiators

A top meta-skill from hiring research is communication, especially:

• asking the right questions,
• giving/receiving feedback without personal conflict,
• explaining work to technical and non-technical stakeholders.

They also argue developers must increasingly understand:

• business problems
• users’ concerns
• and the deep “why” (not just “how”)

This aligns with a shift from hiring generic “software developers” toward product development roles.

Teaching approach at Code/ee-code: ownership-driven learning

Their curriculum philosophy emphasizes:

• providing big problems and concepts
• directing students to find/read materials themselves
• avoiding making a specific instructor-delivered technology the main mechanism

They frame success as requiring ongoing responsibility for self-development.

AI hype doesn’t remove the need for developers

Even if AI accelerates drafting and review, companies still expect developers to:

• navigate specs,
• validate correctness,
• manage tradeoffs (quality vs speed),
• make engineering decisions themselves.

They criticize “vibe coding” hype as creating false expectations.

Managerial implications (quality–speed tradeoffs get harder with AI)

AI can intensify expectations of speed and efficiency, which may lead to:

• measuring output too early.

Recommended management response:

• clarify quality vs speed priorities,
• support learning phases rather than prematurely judging productivity.

Practical recommendations for newcomers

A recurring piece of advice:

• find a problem you’re passionate about and solve it—learning through effort matters even more than the final solution.

Another metaphor:

• don’t just “notice signs”—look for them: cultivate a mindset of what you can solve and act on opportunities, even with low-cost AI experiments/subscriptions.

Key speakers / sources (as named in the subtitles)

• Priit Liivak (host)
• Karmen Tigas (talent manager; leads the market/curriculum discussion)
• Ivo Kund (co-creator of the ee-code AI skills curriculum; long-time software/AI startup background)