Sign in Subscribe
7 min read

🦜 Predictability vs Outcomes, Team Collapse, Lying Roadmaps, DRIs for AI, Defensive Feedback, Lazy Descoping, Slights at Work: TMW #504

Yeah, AI code reviews aren't perfect...
🧑‍🏫
This week's TMW is sponsored by YLD - a software engineering and design consultancy, helping your team succeed by fostering a progressive culture of learning.

Hello, hello, hello, welcome to the week!

Do AI Code Reviews take it a little too far...?

This week, a community member asked how others deal with AI code reviews that flag very unlikely edge cases. The question started the most active conversation in the community in the last week, with many members agreeing that an AI reviewer always finds a problem, because it does not know what “good enough” means for the business. Some teams saw review loops, where two AI tools changed the same code again and again for problems that will probably never occur. Other members said that the problem is larger than the review itself. When AI writes the code, nobody records which decisions were deliberate and which were only assumptions, so errors stay hidden until late.

Members shared some practical methods to control the review:

  • Tell the reviewer to rank each finding by severity, and do not fix the minor findings.
  • Use one agent to find a problem and a different agent to confirm it, so that false findings do not block a merge.
  • Record the intent of each decision next to the code, and send changes with a high risk to a human.
  • For large work, remove all assumptions with a detailed plan before the AI starts, and keep each change small.

If AI now reviews most of your code, who in your organisation decides what “good enough” means, and where is that decision written down?

That's it, let's look at some links! See you next week

Andy @ CTO Craft

CTO Craft Con - Dates for Your Diary

SPONSORED
CTA Image

Navigating the AI Speed Trap: Join Tech Leaders at YLD Innovate on 3 December

We’re excited to invite the CTO Craft community to YLD Innovate, YLD’s exclusive community event for technology leaders, on Thursday 3 December at the Financial Times London HQ. The evening brings engineering, AI, data and product design leaders together for peer discussion, in a closed environment with no sales pitches or vendor presentations.

The evening opens with a fireside chat hosted by Luis Klefsjö, YLD’s Head of Product Design, with Dr Emmanuelle Savarit, an adviser on AI strategy, governance and ethics, author of Practical User Research, with over 20 years’ experience leading research and insight teams, at The London Stock Exchange and Just Eat amongst others. 

Sérgio Ramos, YLD’s Head of Engineering, then hosts a panel with Karl Oskar Teien, Chief Product Officer at the Financial Times, Carl Anderson, VP Engineering at Qonto, and Assad Hussain, Programme Manager at Nationwide. Together they’ll look at regulatory pressure and the trade-off between speed and quality, in decisions that often have no clear owner.

Sign up now - places are limited 👉

CTO Craft Mixers Coming Up

See the whole calendar here - not in your city yet? Drop us a line!

Reads of the Week

Predictability vs Outcomes | Silicon Valley Product Group
A partnership dedicated to teaching best practices to product teams and product leaders
Silicon Valley Product GroupMarty Cagan
Team Collapsing? 6 Warning Signs That Don't Look Like Warning Signs | David Burkus
Is your team collapsing? Here are the warning signs and how to turn them around.
David BurkusDavid Burkus

Leadership, Strategy & Business

Management in the Age of AI
Every few months someone declares that management is dead.
Stay SaaSyStay SaaSy
The Roadmap Is a Lie
5 Questions Before You Commit Engineering Capacity
The Socratic LeadershipPetros Bountis
Best Interview Questions for Assessing Leadership
Discover 15 leadership interview questions and example answers to identify strong leaders. Learn what to look for and how to assess leadership potential.
Metaview BuildsStephanie Tsimis
"You didn't do it right"
Four sincere attempts at OKRs, four failures, and every time: "you didn't do it right." Maybe the framework doesn't fit you -- so how do you find one that does?
A Smart BearJason Cohen

Culture, People & Teams

Put a name on it
Revisiting Directly Responsible Individuals in the AI era
The Engineering ManagerJames Stanier
How to Give Feedback Without Making Someone Defensive
Learn how to give clear, useful feedback without creating unnecessary defensiveness. Five practical ways to reduce threat and keep the conversation open.
Andi Roberts - Executive Coach | Leadership Trainer | FacilitatorAndi Roberts
A Management Minute: Train People on Skills They Need. Period. - Johanna Rothman
I subscribe to HBR’s Management Tip of the Day. A recent tip was “Train Employees on Skills That Evolve with AI.” That’s a lovely sentiment. And totally wrong-headed and insufficient for any workforce, never mind people who might need to learn how to use AI. Yes, I agree that people need training to use AI […]
Johanna RothmanJohanna
7 Rules for Agile Teamwork Without Slowing Everyone Down
Use seven practical rules to help agile teams collaborate, make decisions, share ownership, and finish work without creating unnecessary delays.
Mountain Goat SoftwareMike Cohn

Technology, Operations & Delivery

38 things we've learned about building AI products
Lessons from 4 years of building AI products at PostHog
build modeIan Vanagas
You'll Know in Two Quarters — Leadership Garden
Why "just ship faster" stops working when feedback takes months, and what engineering directors should measure instead.
Leadership GardenCsaba Okrona
Descoping is the Lazy and Incompetent Approach to Meeting Timelines
The Four Abstraction Levels of Refinement: Part 2
Maarten’s NewsletterMaarten Dalmijn
SOC 2 vs. ISO 27001: Which One Does Your Company Actually Need?
If you are a growing technology company and a customer suddenly asks, “Do you have SOC 2 or ISO 27001?”, it is easy to assume they are basically asking for the same thing. They are not. Both framew…
Ido Greengreenido

Stress, Wellbeing & Growth

Slight or Signal
Learning the Game at Work
It Depends: Lessons in Technology LeadershipKevin Goldsmith
You Can't Change Your Boss - Jacob Kaplan-Moss
It’s best to treat your boss’s behavior as fundamentally immutable, like a force of nature. You can certainly try asking the rain to hold off until after your hike, but it’s unlikely to be effective. Assume that trying to get your boss to change their behavior will be similarly ineffective. Your only real choices when you have a bad boss are: learn to live with it, or find another job. (These choices suck, I’m aware.)
Jacob Kaplan-Moss
The unreasonable man - working with CEOs and founders - Doug Rathbone
“All progress depends on the unreasonable man,” wrote George Bernard Shaw, and anyone who has sat beside a founder knows exactly who he meant. The asks are a…
Doug RathboneDoug Rathbone
Hot take
How our judgement shows our perception of reality and how to take control
Alex Ewerlöf NotesAlex Ewerlöf

That’s it!

If you’d like to be considered for the free CTO Craft Community, fill in your details here, and we’ll be in touch!

Campus by CTO Craft: The Go-to Resource for Senior Technology Leaders
Created for CTOs, by CTOs: A transformative learning community for senior technology leaders in startups and scale-ups
CTO Craft Campus

Interested in sponsoring TMW or any of our events? Check out some stats here

Please do remember to share this link if you know of anyone who’d like to receive TMW: https://techmanagerweekly.com

Have an amazing week!

Andy

Published by:

Andy Skipper