This week’s eleven picks all circle back to some assumption that stopped working:
a review process built for a different era, a leadership style forced onto the wrong person, a hiring model chasing the wrong background, a career framework treating management like a promotion instead of a different job entirely.
None of it is really about AI, even where AI shows up. It’s about people testing an inherited default against what’s actually in front of them, and being willing to build something else when it doesn’t hold.
Google’s retirement of Gemini Code Assist triggered a shift from traditional code reviews to proactive commit interventions using Google Jules. Unlike its predecessor, Jules enhances actionability by identifying architectural flaws and applying changes directly in a pull request, streamlining the review process.
Robert Sahlin outlines two key automated processes: an event-driven review for new pull requests and a nightly security check for vulnerabilities. These workflows ensure code clarity, concurrency safety, and security compliance by addressing potential issues before they escalate.
Incorporating tools like Jules shifts code review from passive commentary to active problem-solving, which ensures high-quality outputs and addresses security risks early. Engineering leaders should consider this approach to enhance their team’s efficiency and code reliability.
Enhancing code review processes with automation
Maintaining robust security practices in development
Integrating AI-driven tools for software quality assurance
Conflicts are inevitable in any team dynamic, especially within engineering teams. While many view conflict as detrimental, this article introduces strategies to prevent minor disagreements from escalating into major issues.
James Samuel suggests establishing clear policies and decision-making processes to reduce unnecessary disagreements about methodologies and technologies, thus preventing repetitive debates grounded in personal preferences. Samuel emphasizes the importance of recognizing ‘pinches,’ the early signs of dissatisfaction or misunderstanding.
Engineering leaders are encouraged to empower their teams to handle these pinches through constructive feedback and direct communication. This approach requires training and practice but ultimately leads to a more resilient team that focuses on solving technical issues without letting them become personal.
Engineering leaders must recognize that unresolved conflicts can undermine team productivity and morale. By promoting systems to manage conflicts effectively, leaders ensure a collaborative environment where disagreements enhance rather than hinder innovation.
Building skills to preemptively address misunderstandings
Resolving technical disagreements before they become personal
Creating systems within teams to manage inevitable conflicts
Three food delivery giants, DoorDash, Instacart, and Uber Eats, have uniquely integrated LLMs to enhance their search functionalities, addressing the challenge of interpreting user intent beyond traditional keyword search.
Each company faced the shared issue of deciphering user queries to return relevant results but took distinct paths shaped by their existing infrastructures. This edition of ByteByteGo, explores those different paths.
The article deciphers how each company wisely leveraged their pre-existing structures to push boundaries, thus enhancing search efficiency and user satisfaction.
Understanding how leading tech companies effectively deploy LLMs into search systems reveals practical frameworks for engineering leaders seeking to optimize AI across varied environments. This knowledge is crucial for making informed decisions about integrating AI without disrupting existing operations.
Exploring practical AI integration in product systems
Learning how to tailor AI solutions based on existing tech infrastructure
Understanding strategic AI deployments in competitive tech spaces
Roman Nikolaev pushes back on a piece of standard leadership advice: that a good manager has to become well-rounded, equally strong at problem solving, coaching, and technical depth all at once.
He argues most leaders sit closer to one end of a handful of axes, results versus people, detail versus big picture, technical versus organizational, and that fighting your own instincts all day is what makes the job exhausting.
His fix isn’t more self-improvement. Nikolaev makes the case for a different kind of team-building move entirely, one that covers your gaps without asking you to become someone you’re not.
Engineering leaders spend a lot of energy trying to fix their own gaps instead of building a team that covers them. Nikolaev’s framing gives leaders permission to hire for balance instead of trying to become everything themselves.
Figuring out where you sit on the results versus people, detail versus big picture axes
Building a team that covers your own blind spots
Rethinking what “well-rounded leader” should actually mean
Fahim ul Haq argues that engineers who walk into an Anthropic System Design interview with a standard distributed-systems mental model end up with answers that sound right but miss the point. The reason isn’t that fundamentals don’t apply, it’s that the unit of complexity has shifted. It’s no longer a request or a database, it’s the model itself and how it behaves once it’s embedded in a live system.
That shift changes what the interview is actually evaluating. Instead of scoring how well you can deploy a model, Haq walks through what it takes to reason about the specific ways probabilistic systems break: context as a resource you’re constantly rationing, latency and cost that move together instead of independently, and failure modes that don’t show up in traditional monitoring.
Engineering leaders hiring or interviewing at AI-native companies need to understand that model-driven systems demand a different kind of design thinking than the deterministic systems most interview prep still assumes.
Reframing system design around a probabilistic core instead of deterministic logic
Understanding how context, cost, and latency interact differently in LLM systems
Preparing for technical interviews at AI-native companies
Dr Milan Milanović sits down with Ryan Murphy, who spent five years as an engineering manager at Yelp leading the teams behind its purchasing infrastructure, the kind of systems where a mistake shows up on Wall Street. The interview opens with Milan’s own story about a manager calling him “a good resource,” a line that never sat right, and it turns out Murphy had heard something almost identical early in his career. That moment is part of why he eventually left Yelp to build EM Accelerator.
Murphy’s case is that almost nobody gets real management training beyond whatever their company’s HR department hands them, and teams pay the price for managers learning on the job. He gets specific about what actually separates engineers who stall from ones who get advocated for in promotion conversations, and he’s blunt about where he thinks most managers, including his past self, fail the people under them.
Engineering leaders making the jump from IC to manager, or evaluating how they’re training the managers underneath them, get a direct account of where that transition typically breaks down.
What managers actually look for at each level, from junior through senior
Why Murphy calls management a career change rather than a promotion
Testing whether you’d actually be good at management before committing to it
Jeff Dean, a key figure behind TensorFlow and TPU, discusses AI’s trajectory, predicting more automation in machine learning through sophisticated experimentation loops. He highlights how advancements in AI are enabling models to perform complex tasks like coding for longer durations, shifting the traditional limitations.
Dean shares insights from his Google tenure and delves into energy-efficient hardware, emphasizing the importance of reducing latency and energy consumption for AI inference systems.
He further explores areas where startups can innovate, stressing the enduring value of precise problem selection. Dean encourages thoughtful experiments, reminding us that impactful AI stems from foundational reassessments and interdisciplinary perspectives.
For engineering leaders, Jeff Dean’s insights stress the importance of energy-efficient systems and encourage rethinking conventional assumptions—vital for crafting robust and innovative AI-driven solutions.
Exploring automation in machine learning processes
Rethinking hardware design for energy efficiency
Experimenting with unconventional problem-solving in tech
Sarah Wang and Kimberly Tan speak to Jesse Zhang and Ashwin Sreenivas, co-founders of Decagon. They outline Decagon’s journey towards using smaller, open-source AI models to drive down latency while maintaining performance.
This strategic pivot from reliance on large frontier models has led to significant advancements, making these smaller models outperform in specialized tasks by fine-tuning them to specific needs.
An essential takeaway from the conversation is Decagon’s ‘glass box’ approach. Unlike competitors, they empower enterprises to understand and control the AI interactions thoroughly, enhancing transparency and adaptability. This model encourages rapid development cycles and aligns closely with customer needs, setting Decagon apart in the competitive landscape.
For engineering leaders, Decagon exemplifies how tailored AI applications can enhance enterprise capabilities without over-reliance on expansive models, offering a pragmatic pathway through real-world business integrations.
Understanding the trade-offs in AI model selection for business processes
Learning how AI can transform enterprise-level customer interactions
Exploring strategies for integrating open-source AI within existing systems
Patrick Collison from Stripe challenges whether the traditional lean startup methods remain effective in an AI-driven world. He questions if startups need to be more ambitious from the outset rather than starting narrow and focused.
Collison touches on the importance of fundamental knowledge even in an AI-heavy landscape, suggesting cognitive cache still holds a premium over AI. He shares personal reflections on balancing tasks between AI and human expertise at Stripe.
For early-stage founders, Collison shares insights from the Stripe story, highlighting how solving real user issues and maintaining production users before a public launch were critical.
As AI reshapes industries, engineering leaders must reevaluate strategy frameworks like the lean startup to stay competitive and harness new opportunities effectively.
Reevaluating strategic startup approaches in an AI-centric world
Balancing human expertise and AI capabilities in decision-making
Understanding market adaptation for technical leaders navigating AI advances
Andrew Zigler sat down with Quentin Anthony, VP of AI Engineering at Zyphra, at AMD’s Advancing AI conference, and came away with an argument for why token-spend leaderboards measure the wrong thing entirely. Anthony works directly at the intersection of hardware and model training, writing kernels alongside AI agents, and he’s blunt about why that vantage point makes him skeptical of treating AI usage itself as a sign of progress.
Anthony also gets specific about where models actually break down when steering an agent through unfamiliar territory, and why Zyphra deliberately hires physicists over computer scientists for the kind of first-principles thinking that matters most when nobody’s written the playbook yet. Zigler pairs that with benchmark data on how differently agentic pull requests actually perform once they leave a developer’s machine.
Engineering leaders under pressure to show AI ROI need a sharper way to separate genuine output from activity that just looks like progress, especially when the two numbers tell very different stories.
Why token consumption is a flawed proxy for engineering value
What it actually takes to steer an AI agent through unfamiliar problems
How to measure whether AI-assisted work actually reaches production
In a coherent organization, different aspects like strategy, structure, and incentives align seamlessly, reducing the need for detailed maps or clarity on every small aspect. The simplicity allows teams to infer details effortlessly, much like flipping through aligned layers in a city map.
However, in less coherent organizations, these layers clash, causing confusion and increased effort to navigate the organizational maze. Even experts in these environments can find themselves surprised and constantly reorienting, highlighting the challenges of mismatched goals and unclear ownership.
John Cutler advises against clinging to maps that display these incoherencies. Instead, leaders should focus on identifying misalignments and refactoring the organizational structure. This approach ensures continuous adaptation without relying solely on documentation or artificial intelligence to conceal fundamental misalignments.
Understanding and addressing organizational incoherencies is vital for leaders to foster adaptive and efficient teams. Engineering leaders must recognize and align critical structures to ensure their team’s long-term success and resilience.
Identifying organizational incoherencies in complex environments
Refactoring team processes to enhance alignment and efficiency
Navigating organizational change with minimal friction
The founders, engineering leaders, and CTOs I talk to are building something real, and almost no one knows about it.
The work is good. What’s missing is the machinery around it: content strategy, advertising, outreach, brand development.
That machinery is what took Thriving In Engineering from a newsletter to client conversations, speaking engagements, podcast appearances, collaborations with other industry leaders. None of it happened on its own. My team built it.
Now they’re working with others in the same position. If that’s where you are:
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.