Curlewis analysis challenges AI productivity claims amid Block and Atlassian layoffs
A recent analysis published on Curlewis.co.nz contends that major tech firms are citing AI-driven efficiency to cut headcount despite mixed evidence on organisational productivity gains.
A technical analysis published on Curlewis.co.nz argues that the software industry has abandoned rigorous outcome-based performance measurement in favour of "vanity volume metrics," such as the percentage of code generated by artificial intelligence. The author contends that while AI adoption is widespread, rigorous studies show mixed or negligible productivity gains at the organisational level, yet major tech companies, including Block and Atlassian, continue to cite AI-driven productivity as the rationale for significant workforce reductions.
The piece suggests these decisions are driven by other factors rather than genuine efficiency gains. The author advocates for maintaining established performance measurement standards, such as DORA metrics, rather than relying on adoption intensity or code volume, arguing that the industry has shifted from measuring actual engineering outcomes to tracking superficial adoption statistics.
Supporting this critique, the author references an NBER survey of approximately 6,000 executives which found that while 69% of firms are actively using AI, roughly nine in ten report no measurable productivity impact. Additionally, an RCT study by Anthropic found that AI-assisted developers scored 17% lower on comprehension of the code they had shipped, with no statistically significant productivity gain.
The analysis points to specific workforce reductions to illustrate the disconnect between metrics and reality. Jack Dorsey cut over 40% of Block’s workforce, affecting more than 4,000 people, in February 2026. Shortly after, Atlassian cut 10% of its workforce, reducing headcount by approximately 1,600 people. The author notes that Dorsey’s announcement that Block’s business was strong and gross profit was growing contradicts the narrative that AI productivity gains necessitated the layoffs.
While acknowledging that AI tools are genuinely useful for daily engineering work, the author warns against using them to justify headcount reductions without evidence of genuine idle capacity. The piece concludes that companies should use battle-tested systems to identify underutilised staff rather than relying on token counts or maturity ladders, which often serve vendor interests more than organisational efficiency.
