dream → teamlead · March 28, 2026
AI Changes Engineering Culture
Research, processes, and the leader's role
/ Dream -> Teamlead 2026
Research, processes, and the leader's role
Research, processes, and the leader's role
How AI affects software development — meta-analysis and engineer survey
Practices, culture, trust
Where AI is already daily.
Where adoption remains rare.
How speed, quality, trust connect.
What turns personal gains into team impact.
50+ studies, four source types
Large-scale surveys of developers and tech leaders.
Telemetry and product/engineering logs.
Field experiments and controlled task studies.
Internal corporate research and interviews.
Survey of engineers and tech leads
Self-assessed productivity, quality, trust impact.
AI usage frequency by SDLC task type.
Variable sample size across questions.
Descriptive analysis without causal conclusions.
How to read the results
Sample reflects study context.
Process changes can affect productivity.
Industry metrics remain inconsistent.
Novelty and adaptation skew trends.
Recurring 2023–2025 research patterns
80%+ growth · 10h saved — Default, but uneven — Daily tool; gains depend on task context.
-1–7% stability/throughput — Coding outpaces delivery — Bottlenecks: review, tests, integration.
verify-first + platform — ROI through process — Trust and role depend on SDLC.
AI4SDLC Survey 2025 signals
58% often/always — Everyday AI tool — Code generation is daily practice.
64% vs 32% — Local effect dominates — Productivity outpaces quality and team throughput.
49% / 60% — Trust and expectations diverge — Low trust; business expects impact.
High-risk tasks still need humans
Mass scenarios
58% often use generation.
Coding and debugging are core.
Rare scenarios
24% use AI for code review.
18% use AI for legacy tasks.
Coding speed ≠ value
64% report productivity growth.
Team effect needs mature process.
DORA 2024: AI adoption reduced throughput.
DORA 2025: throughput up; instability stayed.
Verify, don't trust as engineering norm
Quality
32% note quality improvement.
14% report deterioration.
Context and complexity determine the result.
Trust
49% don't trust AI code.
Only 11% trust it.
Verification culture enables impact.
Main takeaway for team leads
Commit → prod is still long.
Review, tests, releases limit flow.
Documentation becomes context quality.
Winners rebuild the work system.
SE 1.0 → SE 2.0: not a tool swap, but a system rebuild
Role-based SDLC vs agent-based SDLC: how the work loop itself changes
Software Engineering 1.0 (Role-based SDLC)
Idea
Req
Dev
Test
Deploy
Support
Product
Analyst
Developer
QA Engineer
SRE
Support Engineer
Handoff losses
Implemented some AI scenarios within roles
Local optimizations inside
On brownfield projects we test AI scenarios by role
Software Engineering 2.0 (Agent-based SDLC)
Idea
Req
Dev
Test
Deploy
Support
Product
Engineer
Support Engineer
Fewer handoff losses
Faster end-to-end scenarios
On greenfield projects we try agent-based development
Nessy Blaze
Nessy
Carry over scenario learnings
Carry over scenario learnings
Software Engineering 1.0 (Role-based SDLC)
Idea
Req
Dev
Test
Deploy
Support
Product
Teamlead
On brownfield projects we test AI scenarios by role
Software Engineering 2.0 (Agent-based SDLC)
Idea
Req
Dev
Test
Deploy
Support
Product
Teamlead
On greenfield projects we try agent-based development
Google, Meta, AWS, DORA
Google: AI inside engineering loop.
Meta: JIT tests for concrete diffs.
AWS: roles compress into one cycle.
DORA: payoff depends on platform.
Why coding speedup slows delivery
Weak version control → conflicts and rework.
Noisy tests → instability.
Agents open PRs faster.
Foundation: branch protection, CI, signing.
Systematic practice transfer between modes
SE 2.0 → frontier — Testing ground for orchestration, JIT tests, auto-docs.
SE 1.0 → foundation — Mature checkpoints, developer platform, safe delivery, operational reliability.
Practice transfer — 1.0 gives guardrails; 2.0 returns automation.
Next level of compression: not just SDLC changes, but the company model itself
Faster code; slower decisions and handoffs
Layers compress — After SDLC roles, organizational layers shrink too.
New bottleneck — decision latency — Approvals and handoffs become bottlenecks.
Leverage over headcount — Focus shifts to output per employee and platform.
Wider span, fewer levels, platform
Wider span
Meta: AI org ratio reaches 1:50.
Nvidia: ~50 CEO reports.
Strong ICs.
Fewer levels
Amazon: +15% IC/manager ratio.
Microsoft, Google: AI-first tooling.
Intel: faster execution, fewer layers.
Without platform, flatness becomes chaos
Need golden paths, policy, observability.
Otherwise managers drown in exceptions.
Ratios like 1:50 are not templates.
Managers design human-agent loops.
New challenges for leaders in the AI-native process world
Not just task distribution
Before
Distributed tasks.
Controlled deadlines and quality.
Hired for fixed roles.
Now
Designs the human-agent loop.
Adjusts quality gates.
Builds diagnostics and review.
Five responsibility areas
Sets AI autonomy by risk.
Bakes verification into definition of done.
Manages context: ADR, runbooks, docs.
Rebuilds metrics, policy, training.
Risks in AI-native transition
Skill erosion — Coding gets easier; deep skills may degrade.
Illusion of speed — Local speed grows while delivery may stall.
Team trust — Verify must be a norm, not distrust.
What to take with you
AI accelerates artifacts, not delivery.
SE 2.0 amplifies engineering discipline.
Team lead designs the human-agent system.
Measure speed of value delivery.
ai4sdlc-research.space · tellmeabout.tech
AI adoption, DevEx, culture
AI4SDLC Research 2025
DORA AI + State of DevOps
GitHub Octoverse + Copilot docs
Stack Overflow, JetBrains, Atlassian
Please share your feedback — it helps us get better
Ask in chat Yandex for Teamleads
Ask via form
Yandex
If you want to explore the materials mentioned in the talk, head to the "Book Cube" channel — all links are there
Alexander Polomodov, Technical Director & Fellow, T-Technologies
@Book_Cube