Managing teams, org design and process in the AI era.
In the 2010s, engineering managers were explicitly told never to make technical decisions again, but today leaders must re-engage with technical details. Larson explains that getting back into writing software serves as a unique reskilling opportunity for modern middle managers facing a tough environment.
▶ 0:00
The revised rules of engineering management with Will Larson
Organizations start with simple initial wins where AI feels exceptionally fast. Next, teams apply AI to bigger problems, causing failures, bugs, and a loss of trust. Finally, teams build real skills by learning proper guardrails, prompting, and context engineering.
▶ 1:11
How to Get Your Org to Adopt Coding Agents (Without Shipping Garbage) — Eyal Blum, Figma
By leveraging modern post-training techniques and clever fine-tuning data set construction, a smaller model could achieve high effectiveness. With full hindsight, a small talented team might have accomplished this around 2018 or 2019 using just a few GPU boxes.
▶ 1:20
John Schulman on dead ends, scaling RL, and building research institutions
Individual developers universally report feeling more productive and note improvements in codebase quality when using AI tools. However, this individual boost does not automatically translate to organizational stability or better delivery outcomes.
▶ 2:11
AI is making software delivery less stable. DORA's Nathen Harvey on the fix | EVOLVE 2026
Acceptance rate only tracks whether a suggestion was taken, not whether that code actually runs in production or drives business value. Relying on it ignores whether developers had to heavily rewrite or debug the suggested code.
▶ 9:16
Measuring the impact of AI on software engineering – with Laura Tacho
If you are in a meeting where people keep talking over each other, a simple hack is to raise your hand or finger. You might be ignored at first, but participants will gradually catch on and start doing the same, leading to a much more civilized conversation. The most powerful moment is when you put your hand back down if your point becomes irrelevant.
▶ 1:37
Tech Truth: Agile Evolution & the Future of SW Engineering • Martin Fowler & Kent Beck • GOTO 2025
Yahoo predated modern open-source libraries and maintained its own forks like Y Apache, YPHP, and Yinst for package management. Because top leadership manually approved server allocations to drive accountability, it inadvertently created a massive black market where teams hoarded outdated, power-inefficient servers.
▶ 10:35
The revised rules of engineering management with Will Larson
Engineers often lose job satisfaction and burn out when reduced to prompt cycles waiting for AI outputs. Meanwhile, senior engineers holding institutional knowledge become overwhelmed bottlenecks trying to prevent bad agent code from entering the codebase.
▶ 2:57
How to Get Your Org to Adopt Coding Agents (Without Shipping Garbage) — Eyal Blum, Figma
A hands-off approach works well for exploratory research with experienced individual contributors who can discover things independently. Conversely, a hands-on approach with more technical oversight is better suited for goal-oriented execution or teams with less experienced people.
▶ 11:14
John Schulman on dead ends, scaling RL, and building research institutions
AI acts primarily as an organizational amplifier, magnifying both the good practices and existing flaws across the SDLC. Without a strategic focus on underlying systems, localized productivity pockets are simply swallowed by downstream chaos.
▶ 5:18
AI is making software delivery less stable. DORA's Nathen Harvey on the fix | EVOLVE 2026
Data from over 180 companies shows that stack trace analysis and refactoring existing code save more time than midloop code generation. Developers spend significant time reviewing generated code, whereas debugging error outputs provides a direct net positive time savings.
▶ 24:34
Measuring the impact of AI on software engineering – with Laura Tacho
Vibe coding is using AI genies to build ambitious projects that were previously too time-consuming to tackle manually. It lets developers focus on high-level skills like defining a distant vision, choosing what to build, and optimizing feedback loops. While it removes language barriers and allows you to work in unfamiliar languages, you still have to watch out for small-scale details and confidently incorrect AI behavior.
▶ 7:44
Tech Truth: Agile Evolution & the Future of SW Engineering • Martin Fowler & Kent Beck • GOTO 2025
The team made aggressive technical decisions by adopting Cassandra to replace sharded MySQL without proper load testing beforehand. The system suffered from critical performance failures and a memory leak caused by a mutable default parameter in a Python API service.
▶ 20:12
The revised rules of engineering management with Will Larson
Shifting verification left from humans to agents prevents bad code and unlocks major productivity gains. For instance, tools like Playwright and MCP allow agents to autonomously explore and verify code instead of relying on human navigation.
▶ 5:04
How to Get Your Org to Adopt Coding Agents (Without Shipping Garbage) — Eyal Blum, Figma
Value functions currently do not provide significant variance reduction on the specific short-horizon tasks researchers are running RL on, such as RLHF and verifiable rewards. While they are useful for variance reduction in other settings, their lack of impact on current tasks makes them less popular.
▶ 17:02
John Schulman on dead ends, scaling RL, and building research institutions
While individual effectiveness increases alongside AI adoption, software delivery instability rises as well. Delivery instability tracks metrics like the frequency of rollbacks and the volume of unplanned rework required after deployment.
▶ 9:38
AI is making software delivery less stable. DORA's Nathen Harvey on the fix | EVOLVE 2026
Research by Dora reveals that AI accelerates the coding tasks that developers actually enjoy doing. This leaves them with a higher proportion of unpleasurable work like toil, meetings, and administrative duties.
▶ 31:08
Measuring the impact of AI on software engineering – with Laura Tacho
Scaled agile frameworks often come with big costs and risks, and organizations naturally try to slow down change. To make large teams more agile, you should focus on improving the communication path between development teams and their users or customers. Additionally, you should analyze the flow of work, find where queues and bottlenecks are, and systematically cut them in half over time.
▶ 19:32
Tech Truth: Agile Evolution & the Future of SW Engineering • Martin Fowler & Kent Beck • GOTO 2025
In the 2010s, engineering managers were explicitly taught never to make a technical decision again and to let go of technology choices. Today, modern engineering leaders are no longer restricted from technical leadership, and roles like CTO require being deeply involved in technical details. This convergence allows leaders to bridge both managerial and technical domains effectively.
▶ 35:46
The revised rules of engineering management with Will Larson
Once you find a useful agent workflow, encoding it into a deterministic flow saves time and tokens. It reserves LLM usage strictly for reasoning tasks while guaranteeing fast, repeatable results for known processes.
▶ 5:46
How to Get Your Org to Adopt Coding Agents (Without Shipping Garbage) — Eyal Blum, Figma
Co-training generators and verifiers creates a virtuous self-improvement cycle where the model gets better at reasoning as it evaluates itself. Additionally, multi-agent games and zero-sum setups provide automatic curriculums and incentives to solve very hard problems using computationally cheap processes.
▶ 24:41
John Schulman on dead ends, scaling RL, and building research institutions
Engineering talent should pivot from churning out features to investing in quality internal platforms and guardrails. These platforms establish trust, reduce operational friction, and safely enable non-engineers to solve problems.
▶ 12:13
AI is making software delivery less stable. DORA's Nathen Harvey on the fix | EVOLVE 2026
Measuring AI impact requires a holistic framework that tracks utilization, impact, and cost across the lifecycle. Because system data alone cannot capture cognitive load or flow state, combining workflow metrics with developer sentiment surveys is essential.
▶ 13:28
Measuring the impact of AI on software engineering – with Laura Tacho
Augmented coding involves a developer actively driving the process and making critical design decisions while using the AI assistant like an exoskeleton. In contrast, vibe coding occurs when someone with little or no programming skills relies entirely on prompts, accepting outputs without understanding or the ability to fix failures.
▶ 36:13
Tech Truth: Agile Evolution & the Future of SW Engineering • Martin Fowler & Kent Beck • GOTO 2025
ZERP shaped a decade of management bloat where perks, hiring throughput, and employee retention became the primary focus. Because managers were discouraged from making technical decisions, they overindexed on process solutions and headcount. This created bloated organizations and shifted priorities until the environment changed around 2022.
▶ 52:12
The revised rules of engineering management with Will Larson
Instructing agents to follow a test-driven development style yields much better results than writing code before tests. This approach sets a clear goal for the agent to strive toward rather than letting it fit tests to pass already-written code.
▶ 6:29
How to Get Your Org to Adopt Coding Agents (Without Shipping Garbage) — Eyal Blum, Figma
John uses AI models heavily for coding, literature searches, and getting initial feedback on his writing. When he has a vague idea, he writes a paragraph or two and has the model flesh it out, which dramatically accelerates finding relevant literature and open-source libraries.
▶ 27:37
John Schulman on dead ends, scaling RL, and building research institutions
DORA's research outlines seven capabilities that, when paired with AI adoption, produce the best outcomes. Key factors include establishing a clear stance, maintaining healthy data ecosystems, using strong version control, and working in small batches.
▶ 13:08
AI is making software delivery less stable. DORA's Nathen Harvey on the fix | EVOLVE 2026
Engineering leaders are recommitting to clean, well-defined interfaces and APIs between services. Clear boundaries make it easier for agentic models to navigate the codebase, which simultaneously benefits human developers.
▶ 36:34
Measuring the impact of AI on software engineering – with Laura Tacho
Drawing from computer science principles established since von Neumann, developers must implement double-checking and inhibition mechanisms to prevent invalid states. A practical technique is asking the LLM the same question or generating the same code multiple times to diff and compare the results for consistency.
▶ 41:46
Tech Truth: Agile Evolution & the Future of SW Engineering • Martin Fowler & Kent Beck • GOTO 2025
Structured career ladders and rubrics are necessary because research shows they drive fairer decisions than having no rubrics at all. Without them, companies tend to undervalue essential maintenance and glue work while overemphasizing flashy projects. However, rubrics should guide decisions rather than replace managerial judgment during promotions.
▶ 58:06
The revised rules of engineering management with Will Larson
Teams should shift as much verification as possible down to deterministic analysis like linters, compilers, and unit tests. Agents can automatically review architectural standards, leaving human reviewers to focus only on high-level functionality and product intent.
▶ 6:57
How to Get Your Org to Adopt Coding Agents (Without Shipping Garbage) — Eyal Blum, Figma
Engineers and researchers consistently exhibit a systematic bias to underestimate project timelines by a factor of two or three. Applying this heuristic suggests that AGI will likely arrive somewhat further out than standard predictions indicate, similar to how self-driving car timelines expanded.
▶ 45:41
John Schulman on dead ends, scaling RL, and building research institutions
AI makes it easy to generate massive applications or huge pull requests with a single prompt, leading reviewers to rubber-stamp code without reading it. Breaking tasks into small batches ensures proper human feedback and maintains high code standards.
▶ 16:25
AI is making software delivery less stable. DORA's Nathen Harvey on the fix | EVOLVE 2026
Companies should combine system workflow metrics with self-reported developer sentiment surveys to track utilization, impact, and cost. Using a developer experience index helps reveal broader organizational gains like an 11% boost in developer experience.
▶ 43:02
Measuring the impact of AI on software engineering – with Laura Tacho
Junior developers should resist the pressure to immediately start the next feature and instead use the gaps to learn and master their craft. They can leverage the AI as an endless tutor to explore alternative implementations, run coverage testing, find missing tests, and refactor code.
▶ 51:16
Tech Truth: Agile Evolution & the Future of SW Engineering • Martin Fowler & Kent Beck • GOTO 2025
Instead of adding headcount to a late project, which destroys small ownership pods, managers should keep durable high-ownership teams focused on the problem. They can ask for short-term commitments and extra hours from the team while removing bureaucratic distractions. This prevents the constant organization-wide panic and burnout seen in reactive staffing surges.
▶ 1:17:18
The revised rules of engineering management with Will Larson
Spending up to a week writing a comprehensive implementation plan allows developers to make and flush out decisions before sending it to the agent. This approach reintroduces the intellectual craft and joy of building back into the development process.
▶ 7:50
How to Get Your Org to Adopt Coding Agents (Without Shipping Garbage) — Eyal Blum, Figma
Tinker is a low-level fine-tuning API that provides a small set of primitives for training and sampling without requiring users to manage GPUs or distributed systems. It is currently designed for sophisticated ML practitioners who want to write custom post-training algorithms, though higher-level tooling is planned over time.
▶ 47:58
John Schulman on dead ends, scaling RL, and building research institutions
Adopting any new technology introduces a productivity J-curve where output dips due to learning curves and verification taxes before recovering. Leadership must anticipate this dip rather than expecting immediate ten-fold gains from handing out AI licenses.
▶ 25:32
AI is making software delivery less stable. DORA's Nathen Harvey on the fix | EVOLVE 2026
AI makes it trivially easy to write large changes all at once, leading to bigger batch sizes and increased risk. Research predicts that a 25% increase in AI adoption could result in a 7.2% reduction in delivery stability.
▶ 53:50
Measuring the impact of AI on software engineering – with Laura Tacho
Businesses dealing with intricate domain contexts, strict regulations, and extensive audit requirements retain strong defensibility despite new AI tools. The difficulty lies not just in writing code, but in handling deeply specific rules, third-party relationships, and legal edge cases. These details require deep domain knowledge that generic AI harnesses struggle to capture automatically.
▶ 1:11:53
The revised rules of engineering management with Will Larson
A strong plan starts with a clear executive summary to prevent agent drift and keep the end goal in sight. It must also be broken down into small, independently verifiable phases that correspond to manageable pull requests.
▶ 8:41
How to Get Your Org to Adopt Coding Agents (Without Shipping Garbage) — Eyal Blum, Figma
The industry approached AI backward by focusing on feature generation before code review. Using AI to review code changes first is a safer starting point that helps teams understand model capabilities and build trust before tackling riskier generative tasks.
▶ 27:49
AI is making software delivery less stable. DORA's Nathen Harvey on the fix | EVOLVE 2026
Highly regulated industries like financial, insurance, and pharma are experiencing the best results. Their success stems from being deliberate and structured in their rollouts, managing acceptable use policies and security intentionally.
▶ 59:16
Measuring the impact of AI on software engineering – with Laura Tacho
Effective executives need both deep technical understanding and strong bureaucratic skills. Traditional middle management often selected for bureaucratic navigation while losing touch with technology, whereas line managers retain better technical instincts. A modern executive must master both to structure functions and allocate effective processes.
▶ 1:27:11
The revised rules of engineering management with Will Larson
Instead of forcing skeptics to use AI, organizations should bring them in and take their critical feedback seriously. Their skepticism highlights real tool failures and validation gaps, making their feedback the ideal roadmap for making AI safe.
▶ 11:53
How to Get Your Org to Adopt Coding Agents (Without Shipping Garbage) — Eyal Blum, Figma
Middle managers face a stark choice between reskilling or risking irrelevance as their roles are flattened. To adapt, leaders can take lateral or backwards steps into line management to get back into the details of building software. Will Larson exemplifies this by moving from leading 250 people at Carta to a 25-person team at Imprint.
▶ 1:33:52
The revised rules of engineering management with Will Larson
Because human attention is a scarce resource in an AI-heavy workplace, teams must clearly distinguish between human-written and AI-generated content. Marking your intent allows readers to know how much scrutiny and attention a message requires.
▶ 12:58
How to Get Your Org to Adopt Coding Agents (Without Shipping Garbage) — Eyal Blum, Figma
High-volume AI writing lacks genuine perspective and fails to drive real industry impact. Building lasting leverage requires spending significant time developing deep, memorable ideas rather than churning out content mills. Writers who rely on AI slop risk destroying decades of carefully built professional reputation.
▶ 1:44:51
The revised rules of engineering management with Will Larson
No answer mentions that.