John Adib
John Adib
Back to talks
TechLead Conf logo

Panel · TechLead Conf London 2026

Hands-on Engineering Leadership in the AI Era

When
October 23, 2026
Where
The Brewery, London
Organizer
TechLead Conf

I returned to TechLead Conf for the second time, this year as a panelist across four discussions and moderator of the panel on AI-native junior engineers.

The 2026 edition centred on hands-on AI adoption across the software development lifecycle, from discovery to production. My four panels focused on the people and decisions behind that transition: how engineers develop experience, which skills matter when hiring, how teams measure the impact of AI, and how engineering leadership itself changes.

Four panels. Four perspectives on AI.

Moderator and panelist

Skipping the Struggle: Can AI-Native Juniors Become Seniors?

AI speeds up implementation and exposes juniors to more problems. Progression still depends on judgment, understanding what they ship and owning the outcome.

  • Seniority is ownership under ambiguity: breaking down a loosely defined problem and planning for failure and recovery, well beyond writing code.
  • Copying generated code builds output, not capability. Asking engineers to explain the approach, alternatives and failure modes shows whether they learned.
  • Some experience only arrives after delivery, when customers use the product in ways nobody expected. Faster building creates more chances to learn, not a shortcut past production.
  • Hiring only seniors risks inflated titles and thin promotion paths. Teaching juniors also makes experienced engineers examine what they really know.

My part

I moderated the panel and shared how my team runs three AI reviewers on every pull request, with human review when needed. That raised the question the room kept returning to: how do juniors keep learning when code review is no longer a routine point of human contact?

Alongside Bogdana-Elena Avadanei, Dünya Kirkali, Dmitry Kmita, Mostafa Hadian Dehkordi, Semih Sezer.

Panelist

AI-Era Hiring: What Skills Last?

An interview should show how a candidate frames a problem, makes decisions and verifies the result with the tools they would use at work. Tool fluency counts, alongside engineering depth and ownership of delivery.

  • Speed and syntax accuracy are weaker signals than before. Judgment, design and technical depth still decide hires.
  • Poor requirements now produce the wrong thing faster, so engineers increasingly own the work from requirements to maintenance.
  • Juniors need bounded but complete problems: enough context to see why it matters, room to decide and a clear way to verify.
  • Changing an interview moves where candidates drop out. Track stage progression and hiring outcomes, not pass rates alone.

My part

I walked through the live interview I designed to replace our take-home exercise. Candidates get a small repository and about ten bugs and features in advance, then work through them with an AI tool they know. One reported bug is deliberately false, and the signal is whether they validate it before fixing it. I also argued that understanding systems and product value outlasts expertise in any one language, after building and publishing a native Swift app in a weekend with no prior Swift experience.

Alongside Bogdana-Elena Avadanei, Valerii Iatsko, Nihan Bircan, David Saidon, Rahul Kumar, Drishti Agarwal, Muhammad Shuaib Anjum, Tom Kadwill.

Panelist

Measuring AI/Agent Impact: KPIs, ROI, Real Productivity

There is no single reliable productivity number. Adoption, delivery speed, quality, spend and customer impact have to be read together, because speeding up one stage often just builds a queue at the next.

  • Installs, active users and tokens show access, not value. Token use can mean a useful task or an agent going in circles.
  • The bottleneck moves: faster implementation exposes review, then testing, then product decisions.
  • Review intent and specifications before code is generated, then verify the delivered system against them.
  • Spend alone shows neither skill nor value. Read it alongside outcomes.

My part

I argued for outcomes over output. Pull request volume on my team has grown several times over, but the result that matters is hitting quarterly objectives, and our constraint has moved from implementation to prioritisation and sign-off. I also explained why we treat AI budgets as an experiment, adjusting caps as we learn how teams use the tools.

Alongside Kseniia Korostelova, Valerii Iatsko, Dmitry Kmita, Alexandru-Daniel Tufa, Mostafa Hadian Dehkordi, Tom Kadwill, Gergana Ivanova.

Panelist

TechLead Unconference: Engineering Leadership in the Age of AI

AI can raise individual output while weakening collaboration and adding stress. Leaders have to design how people align, learn and share ownership. Giving everyone an agent does not create a team.

  • Cheap code makes review expensive. Align on intent and architecture early, then ship smaller, sequenced pull requests.
  • Specifications cannot remove discovery. Estimate how hard a change is to validate, not how hard it is to write.
  • Stress comes from job fears, rising expectations and constant tool change. Leaders need to offer time to learn and a credible path forward.
  • Automations built by finance, supply chain or marketing teams need owners, documentation and recovery steps once the business depends on them.

My part

I shared how a long-time Python engineer on my team grew into frontend through small tasks and steady feedback, later rebuilding an interface of more than 50 pages on our design system. I also described our agent-assisted delivery flow: an issue triggers an investigation and report, a person decides whether to pursue it, then agents implement, attach evidence and open a pull request for a human to test and merge.

Alongside Kevin Ball, Patrick Akil, Fabio Dias.

Held on 23 October 2026 at The Brewery, London. Lineups follow the official programme. Recaps come from my notes on each recording.

My panel at TechLead Conf 2025

Resources

Event page