Six engineers, 76 days: hiring has become an infrastructure decision
AWS rebuilt a critical part of Bedrock with six engineers instead of thirty. When a team is six people, one wrong hire is a sixth of its capacity.
Thirty engineers for 12 to 18 months. Or six engineers for 76 days.
That's the choice an AWS team made to rebuild a critical part of Bedrock's inference engine, the one that routes billions of requests to the models.
The conventional estimate said 30 engineers, 12 to 18 months. Six experienced engineers did it in 76 days with coding agents, and AWS talks about individual productivity multiplied by 20, achieved by changing how they work, not just the tools.
What interests me isn't the productivity figure. It's what disappears along with the 24 roles.
Thirty people means several teams, managers, meetings, coordination. Six people is a group assembled around a problem, who solve it and move on to the next. Amazon has been working on this since 2024: Andy Jassy asked to raise the ratio of individual contributors per manager by at least 15%.
For hiring, it changes the equation.
In a team of thirty, an average hire gets diluted. In a team of six, it's a sixth of the capacity. Getting one person wrong is no longer an HR cost. It's an infrastructure risk.
And AWS's six weren't six promising juniors. They were six experienced engineers, chosen for it.
