Why Agentic Coding Amplifies Whatever Your Team Already Is

Presented at Agentic Shift in Munich on June 2026

Agentic coding doesn't just change how you write code. It changes where the bottleneck sits. It used to be implementation. Now it moves to other parts: maybe product discovery (are your specs clear enough for an agent to act on?) or code validation (how do you review code no human wrote?) .. or maybe your code quality.

​When teams adopt agentic workflows, these things become hard to ignore. In this talk I'll walk through what actually happens in that transition:

​Which existing practices tend to break first, and which new bottlenecks show upWhich practices hold up for collaboration and for code quality

​This talk helps you figure out where your team is in adopting agentic engineering and what to start with, both for your personal and team

Transcript

1. WHY AGENTIC CODING AMPLIFIES WHATEVER YOUR TEAM ALREADY IS What it really takes for AI to truly be helpful. TEREZA IOFCIU Agentic Shift - June 2026
2. A SENIOR PERSON JOINS YOUR TEAM TODAY. Strong. Experienced. Exactly who you wanted.
3. WHAT DO THEY NEED TO BE PRODUCTIVE ON DAY ONE? How long until they really are? Two weeks? Two months? A quarter?
4. THE ONBOARDING TIMELINE DAY 1 ACTUALLY PRODUCTIVE months later MOST OF WHAT THEY NEED IS NOT WRITTEN DOWN ANYWHERE.
5. WHAT NEW COMERS HAVE TO ABSORB How we decide what is worth doing What "good" actually looks like here Who to ask, and what to trust What "done" means IT LIVES IN PEOPLE'S HEADS. SO ONBOARDING TAKES A QUARTER.
6. THE CONTEXT YOU'D GIVE THAT PERSON IS THE CONTEXT YOU'D GIVE AN AGENT. And it needs to be written down, because the machine will not absorb it by osmosis. Of course the AI workflows are getting better… more questions are asked, work gets planned & feedback is asked… but they still make assumptions in unexpected ways
7. WHERE THE BOTTLENECK MOVES NEXT 1 Product discovery: are your specs clear enough for an agent to act on? 2 Code validation: how do you review code no human wrote? 3 Code quality: is your codebase a good enough example for AI to copy?
8. Hi, I’m Tereza
9. 1 WHY AI HELPS YOU, NOT THE ORG
10. ENGINEERING DID GET FASTER Not 10x. Closer to 20-30% once you wire up the right tools. Agents take the task research and the planning They do the prototyping and the review prep People spend more time on the parts they like THE WIN IS REAL. AND IT STOPS AT THE EDGE OF THE TEAM.
11. HOW AI PRODUCTIVITY FEELS LIKE. INDIVIDUAL ORG https://www.mindstudio.ai/blog/ai-adoption-gap- employees-utilization-2026 INDIVIDUAL THROUGHPUT UP. ORG-LEVEL VALUE FLAT.
12. WHAT DID NOT SPEED UP The part of the org that decides what is worth building Most teams feel it as: product discovery is slow It runs on shared understanding that lives in heads AI DID NOT BREAK THIS. IT REVEALED IT.
13. WE GAVE THE CAR A BIGGER ENGINE ENGINE engine = agentic coding SPEED WITHOUT DIRECTION IS A FASTER WRONG TURN.
14. AND THE HUMAN IN THE LOOP IS TIRED Thirty AI pull requests every morning, each a judgment call You start far more than you can ever finish You are the quality gate for far more output YOU CANNOT HAND-REVIEW YOUR WAY OUT OF THIS. https://pydantic.dev/articles/the-human-in-the- loop-is-tired
15. 2 WHAT WE NEED TO WRITE DOWN
16. WRITE DOWN WHAT GOOD WORK LOOKS LIKE. The single thing that makes working with AI better is being explicit about expectations.
17. IN ENGINEERING IT ALREADY HAS NAMES Definition of ready, definition of done The developer workflow Testing and code quality expectations BE SPECIFIC: WRITE DOWN WHAT FINISHED AND GOOD MEANS.
18. NOT A RULEBOOK. EXAMPLES. GOOD – Eng: a ticket that names the user and the test for done – Brief: who it is for, what changes, how we will know it worked NOT GOOD – Eng: "make it better”, “it works” – Brief: "we should look into this" This is how LLMs and Agents are optimized
19. YOU MAY ALREADY HAVE IT WRITTEN DOWN CLEAN CODE + CURRENT DOCS AI copies a good example. The output is good. TECH DEBT + STALE DOCS AI amplifies the mess. Slop in, slop out. https://www.pragmaticcoders.com/blog/is- clean-code-dead-why-ai-actually-makes-good- coding-more-important-than-ever AI AMPLIFIES WHATEVER YOUR TEAM ALREADY IS.
20. CODIFY YOUR EXPECTATIONS IN ENGINEERING – Specs – Skills – Templates IN THE REST OF THE ORG – What a good brief looks like – What a good summary looks like – Packed into shared skills, templates, workflows
21. WE ARE ABOUT TO ONBOARD THE AI BETTER THAN WE ONBOARDED PEOPLE.
22. AND THE HUMANS INHERIT IT We never invested in this clarity for people Now we write it down because the agents force us to Clearer expectations, faster onboarding, less guessing AI is great at writing & maintaining documentation
23. 3 OUT OF ENGINEERING, INTO THE ORG
24. AIIFY ONE TEAM TASK AT A TIME AIify (verb): find the task that repeats, research, aggregate, deliver in a set format, and build it with the team once. A SUBSCRIPTION IS A GOOD START. HELP PEOPLE DO THE TASK.
25. INDIVIDUAL, TO TEAM, TO ORG INDIVIDUAL TEAM ORG AI ambassadors carry the working patterns between teams, so each person does not rediscover them. SHARE THE WORKFLOW THE MOMENT A SECOND PERSON NEEDS IT.
26. AND CREATE THE PULL Bring the excitement: hackathons and demos, not mandates Set up spaces where people can safely experiment The more people try, the more ideas they get for where AI fits EXPERIMENTATION CREATES DEMAND. MANDATES CREATE COMPLIANCE.
27. WHERE IS YOUR TEAM RIGHT NOW? Individuals quietly experimenting, nothing shared yet A team sharing a few workflows, but good still lives in heads The org codifying what good looks like, inherited by default START WHERE YOU ARE: PICK ONE TASK, DEFINE GOOD, RUN THE LOOP.
28. TREAT AI ADOPTION AS AN ITERATIVE EXPERIMENT OBSERVE MEASURE IMPROVE FOR YOU YOUR TEAM THE ORG
29. Models and tools will keep changing. What good looks like changes far more slowly. Write that down, and every new model just gets better at hitting a target you already own.
30. EVERYONE WANTS MORE PRODUCTIVITY. ALMOST NO ONE MEASURES WHAT GOOD IS. How would you measure your current productivity?
31. Thank you. Tereza Iofciu terezaiofciu.com Connect on LinkedIn → Keep on trying things out & share your experiments!
32. REFERENCES • DORA, State of AI-assisted Software Development 2025 - dora.dev/dora-report-2025 • Faros AI, The AI Productivity Paradox 2025 - faros.ai/blog/ai-software-engineering • Faros AI, Acceleration Whiplash 2026 - faros.ai/research/ai-acceleration-whiplash • GitClear, AI Copilot Code Quality 2025 - gitclear.com/ai_assistant_code_quality_2025_research • Pydantic (L. Summers), The Human in the Loop is Tired 2026 - pydantic.dev/articles/the-human-in-the- loop-is-tired • Anthropic, Demystifying evals for AI agents - anthropic.com/engineering/demystifying-evals-for-ai-agents • Berkeley Haas / HBR, AI doesn't reduce work, it intensifies it 2026 • Tom DeMarco, Slack: Getting Past Burnout, Busywork, and the Myth of Total Efficiency (2001) • Blog post on loops https://alexlavaee.me/blog/production-agent-loops/