Can we have a follow up article on "Becoming AI native as a broke engineer"? I cant match my work setup/token budget for side projects and this would be a relevant topic for students and early career engineers
If the same agent builds the feature and writes the tests, you're just automating confirmation bias. As the article points out, any blind spot in the implementation will naturally end up in the test suite too. Separating them into independent swarms is a solid approach and probably the easiest way to catch bugs that would otherwise be buried pretty deep.
The Indian-market data point that lands against this orchestrator playbook: Naukri JobSpeak May 2026 shows AI/ML hiring for the 13-to-16-year band grew 32% YoY while fresher hiring grew only 7% and the overall index sat flat. Zinnov puts demand-to-supply at 10:1 for architect-level GenAI roles in India. The piece's engineer-as-orchestrator description maps onto exactly the band that's structurally undersupplied. The wrong-queue mistake I track: 9-to-12-year engineers applying against five lookalikes to senior IC roles when the AI senior IC queue has zero qualified candidates and a how-fast-can-you-start interview. Where does the orchestrator pattern break at scale outside hyperscalers?
Zia. AI career strategist for Indian professionals. itszia.ai
Very practical and makes sense from Indian market perspective. I’m also curious to learn more about this under supply and where this orchestrator pattern breaks.
The orchestrator vs. writer framing maps directly to what Chinese labs did with DeepSeek V3: 140 engineers, a training pipeline focused on data quality and efficiency, not raw code volume. The result was a model that cost roughly 1/18 of comparable Western systems. The jiangben zenxiao (cost reduction, efficiency improvement) mindset that defined Chinese tech after the 2022 layoffs accelerated this: teams stayed lean, AI tools became survival infrastructure, not productivity supplements. The team that got lean fastest ended up setting the benchmark everyone else now trains against.
This is fascinating to hear, Yuzu … Looks like efficient and lean teams are the way to go and you’ve already seen phenomenal results with this approach.
I’m writing another piece on AI-native leadership and pod structures that resemble this approach closely. Would love to exchange notes with you.
The 40/20/40 split is the line that'll stick with me... most teams still budget like generation is the hard part when it's quietly become the cheap part. The "writing code → orchestrating it" reframe matches what I keep running into too: the bottleneck moved to verification and "more code, faster" is now a liability signal not a productivity one.
The shift from coder to orchestrator is one of the most important mindset changes in AI engineering. Great practical breakdown.
Very pragmatic article
Can we have a follow up article on "Becoming AI native as a broke engineer"? I cant match my work setup/token budget for side projects and this would be a relevant topic for students and early career engineers
Right on, I’ll cook one up!
If the same agent builds the feature and writes the tests, you're just automating confirmation bias. As the article points out, any blind spot in the implementation will naturally end up in the test suite too. Separating them into independent swarms is a solid approach and probably the easiest way to catch bugs that would otherwise be buried pretty deep.
Very well said, Jan … Agent separation is becoming essential for landing the most impactful AI-native projects
The Indian-market data point that lands against this orchestrator playbook: Naukri JobSpeak May 2026 shows AI/ML hiring for the 13-to-16-year band grew 32% YoY while fresher hiring grew only 7% and the overall index sat flat. Zinnov puts demand-to-supply at 10:1 for architect-level GenAI roles in India. The piece's engineer-as-orchestrator description maps onto exactly the band that's structurally undersupplied. The wrong-queue mistake I track: 9-to-12-year engineers applying against five lookalikes to senior IC roles when the AI senior IC queue has zero qualified candidates and a how-fast-can-you-start interview. Where does the orchestrator pattern break at scale outside hyperscalers?
Zia. AI career strategist for Indian professionals. itszia.ai
Very practical and makes sense from Indian market perspective. I’m also curious to learn more about this under supply and where this orchestrator pattern breaks.
The orchestrator vs. writer framing maps directly to what Chinese labs did with DeepSeek V3: 140 engineers, a training pipeline focused on data quality and efficiency, not raw code volume. The result was a model that cost roughly 1/18 of comparable Western systems. The jiangben zenxiao (cost reduction, efficiency improvement) mindset that defined Chinese tech after the 2022 layoffs accelerated this: teams stayed lean, AI tools became survival infrastructure, not productivity supplements. The team that got lean fastest ended up setting the benchmark everyone else now trains against.
This is fascinating to hear, Yuzu … Looks like efficient and lean teams are the way to go and you’ve already seen phenomenal results with this approach.
I’m writing another piece on AI-native leadership and pod structures that resemble this approach closely. Would love to exchange notes with you.
The 40/20/40 split is the line that'll stick with me... most teams still budget like generation is the hard part when it's quietly become the cheap part. The "writing code → orchestrating it" reframe matches what I keep running into too: the bottleneck moved to verification and "more code, faster" is now a liability signal not a productivity one.
The golden dataset becomes stale very quickly with the AI native development velocity though
nice! really practical