Software Lead Weekly 
Issue #719, 4th September 2026
This Week's Favorite
The Harness, the Horse, or the Hay
14 minutes read.
Brett Queener draws the line that decides which infrastructure survives the AI labs eating the stack: "Anything that gets the median developer working faster increases inference, so they'll ship a free version and thank you for the roadmap. Anything that reduces inference, makes the customer portable, or makes somebody accountable outside the relationship, they'll never build well." -- Worth running your own product roadmap through that test before a lab does it for you.
Read it later via Instapaper.
Share it via Twitter or email.
14 minutes read.
Brett Queener draws the line that decides which infrastructure survives the AI labs eating the stack: "Anything that gets the median developer working faster increases inference, so they'll ship a free version and thank you for the roadmap. Anything that reduces inference, makes the customer portable, or makes somebody accountable outside the relationship, they'll never build well." -- Worth running your own product roadmap through that test before a lab does it for you.
Read it later via Instapaper.
Share it via Twitter or email.
Culture
Staff Software Engineer Prompting a Cloud Agent to Stop Adding Narrative Copy to the UI
1 minutes read.
My humble effort to help you start the weekend with a smile.
Read it later via Instapaper.
Share it via Twitter or email.
1 minutes read.
My humble effort to help you start the weekend with a smile.
Read it later via Instapaper.
Share it via Twitter or email.
The Incumbents Are Coming
7 minutes read.
Seema Amble with a helpful framing around the hierarchy of value: retrieval, process, policy, principal agents, ranked by how much judgment they require. Most incumbents are still stuck moving from process toward narrow policy, bounded by the record they own. Where does your own AI roadmap actually sit on that ladder, and what would it take to reach judgment-level agents?
Read it later via Instapaper.
Share it via Twitter or email.
7 minutes read.
Seema Amble with a helpful framing around the hierarchy of value: retrieval, process, policy, principal agents, ranked by how much judgment they require. Most incumbents are still stuck moving from process toward narrow policy, bounded by the record they own. Where does your own AI roadmap actually sit on that ladder, and what would it take to reach judgment-level agents?
Read it later via Instapaper.
Share it via Twitter or email.
The Elephant in the Brownfield
5 minutes read.
This is where adoption is so elusive and building products that make you fall in love with them is often an art more than science: "You cannot make people change their ways of working. You can only increase their willingness to change. I got the best work done when I could appeal to that willingness. In those cases, they made my goals theirs. I failed when I could not do that and had to wrestle with escalations instead, because the goals conflicted with what they felt was important."
Read it later via Instapaper.
Share it via Twitter or email.
5 minutes read.
This is where adoption is so elusive and building products that make you fall in love with them is often an art more than science: "You cannot make people change their ways of working. You can only increase their willingness to change. I got the best work done when I could appeal to that willingness. In those cases, they made my goals theirs. I failed when I could not do that and had to wrestle with escalations instead, because the goals conflicted with what they felt was important."
Read it later via Instapaper.
Share it via Twitter or email.
There's No Reason for Software to Be Slow Anymore
11 minutes read.
Dan Luu recounts Jamie Brandon prepping for a performance interview by trying Anthropic's public takehome, then handing it to Claude to finish: "he's a reasonable performance engineer and he got an offer for the performance job he wanted, but on a well-defined optimization problem, he doesn't stand a chance against a decent model" -- the uncomfortable part isn't that AI writes code now, it's that years of hard-won expertise can lose to an agent within an afternoon, as long as the problem is bounded and well-defined. This is the key point, though: the better you can define the problem and set clear boundaries around it (requirements & constraints), the better results you'll get and the fewer "turns" you'll need to spend with your tools. Take more time to plan. Go deeper into understanding the problem.
Read it later via Instapaper.
Share it via Twitter or email.
11 minutes read.
Dan Luu recounts Jamie Brandon prepping for a performance interview by trying Anthropic's public takehome, then handing it to Claude to finish: "he's a reasonable performance engineer and he got an offer for the performance job he wanted, but on a well-defined optimization problem, he doesn't stand a chance against a decent model" -- the uncomfortable part isn't that AI writes code now, it's that years of hard-won expertise can lose to an agent within an afternoon, as long as the problem is bounded and well-defined. This is the key point, though: the better you can define the problem and set clear boundaries around it (requirements & constraints), the better results you'll get and the fewer "turns" you'll need to spend with your tools. Take more time to plan. Go deeper into understanding the problem.
Read it later via Instapaper.
Share it via Twitter or email.
Peopleware
I Think I Accidentally Became a Marketing Engineer
10 minutes read.
"Someone has to define the destination, set the constraints, own the quality gates, and decide what counts as finished. That work cannot be automated away." -- Zach Chmael argues that the people who get squeezed by AI stay in the execution layer, while the ones who compound move into orchestration: encoding judgment, designing the system, delegating labor to agents. Are you spending your week producing, or designing the system that produces? We all need to make that shift before it gets made for us.
Read it later via Instapaper.
Share it via Twitter or email.
10 minutes read.
"Someone has to define the destination, set the constraints, own the quality gates, and decide what counts as finished. That work cannot be automated away." -- Zach Chmael argues that the people who get squeezed by AI stay in the execution layer, while the ones who compound move into orchestration: encoding judgment, designing the system, delegating labor to agents. Are you spending your week producing, or designing the system that produces? We all need to make that shift before it gets made for us.
Read it later via Instapaper.
Share it via Twitter or email.
Grounding Claude in Truth: The Semantic Layer for an AI-first Data Team
4 minutes read.
Louis Ryan found that Claude answered Fin's core business metrics correctly only 65-70% of the time when left to explore the warehouse on its own, close, but not close enough when the output feeds comp or QBRs. Building a semantic layer of approved metric definitions, SQL templates, and known traps pushed that to 100%. What's the equivalent "close enough" number quietly sitting inside your own AI-assisted workflows right now?
Read it later via Instapaper.
Share it via Twitter or email.
4 minutes read.
Louis Ryan found that Claude answered Fin's core business metrics correctly only 65-70% of the time when left to explore the warehouse on its own, close, but not close enough when the output feeds comp or QBRs. Building a semantic layer of approved metric definitions, SQL templates, and known traps pushed that to 100%. What's the equivalent "close enough" number quietly sitting inside your own AI-assisted workflows right now?
Read it later via Instapaper.
Share it via Twitter or email.
Write the Skill Yourself
4 minutes read.
"You get their answers, not the process of arriving at your own. And the process is most of the point." -- Josh Herzig-Marx is talking about installing someone else's Claude Skill, but the same warning applies to borrowing anyone's playbook, prompt library, or onboarding doc wholesale. The value was never the artifact, it was being forced to explain the tacit judgment you never had to spell out before. Do you sort your own tacit knowledge that clearly before you hand it to anyone, human or AI?
Read it later via Instapaper.
Share it via Twitter or email.
4 minutes read.
"You get their answers, not the process of arriving at your own. And the process is most of the point." -- Josh Herzig-Marx is talking about installing someone else's Claude Skill, but the same warning applies to borrowing anyone's playbook, prompt library, or onboarding doc wholesale. The value was never the artifact, it was being forced to explain the tacit judgment you never had to spell out before. Do you sort your own tacit knowledge that clearly before you hand it to anyone, human or AI?
Read it later via Instapaper.
Share it via Twitter or email.
Inspiring Tweets
@christophersaum: A GTM pattern I’m seeing among some of the best young founders: They’re using highly curated dinners to build customer communities. One founder selling into health systems is hosting Michelin dinners for health-system C-suites. Another is convening fintech leaders in New York. It’s targeted, high-trust, and working extremely well.
@johncrickett: We've known this for decades: "Less code means less code to maintain, fewer bugs" About time we started actually internalising it.
- Oren
P.S. Can you share this email? I'd love for more people to experiment and improve their company's culture.
P.S. Can you share this email? I'd love for more people to experiment and improve their company's culture.