Context

By mid-2022, AWS Professional Services EMEA had a serverless problem that looked like a staffing problem. Every engagement that involved Lambda, EventBridge, or Step Functions required one of four people on the team, or it required one of those four to review the architecture before delivery. The four people were good. They were also at capacity, consistently over-allocated, and becoming a single point of failure for a practice that was growing faster than the expertise could.

The waiting list for a serverless engagement was measured in weeks. Customers who wanted to move quickly chose delivery partners instead of AWS PS, because delivery partners could start immediately. The practice was leaving revenue behind every quarter because four people cannot be in multiple places at once.

The real problem

The instinct was to hire more serverless specialists. The actual problem was that the knowledge was not codified anywhere. The four specialists were excellent because they had built up judgment through dozens of engagements. That judgment lived in their heads and did not transfer through job descriptions or pairing. New hires would take eighteen months to reach the same level, and eighteen months was not useful.

The structural problem was that there was no path for a capable generalist consultant to reach working serverless competency in a reasonable timeframe. Not expert-level competency. Working competency: able to design a standard serverless architecture, recognize the failure modes, and deliver an engagement with appropriate specialist oversight rather than specialist dependency.

Approach

The bootcamp design started from the outcome and worked backward. The outcome was a consultant who could take a serverless engagement from discovery through delivery without a specialist in the room, and who knew when to escalate. That defined the curriculum.

The bootcamp ran over four days, residential, with two weeks of remote preparation before arrival. The remote preparation was labs only: hands-on Lambda, SQS, EventBridge, and Step Functions. No lectures, no videos. The goal was to arrive at the residential phase with the mechanics in muscle memory.

The residential days were entirely case-based. Each day used a real engagement from the practice’s history, anonymized, and ran participants through the problem framing, the architecture decision, the failure mode they missed, and what they would do differently. The facilitators were practitioners, not trainers.

Three things the curriculum deliberately excluded: certification paths, vendor presentations, and any material that could have been read in a documentation page. If it was in the docs, it was not in the bootcamp. The bootcamp existed to transfer judgment, which cannot be read.

The assessment at the end was a new customer scenario, unseen, with twenty-four hours to produce a recommendation and defend it in front of two experienced practitioners. Pass or fail. A consultant who passed was cleared to take serverless engagements with a reduced oversight requirement. A consultant who did not pass got specific feedback and could retake after remediation.

What worked

  • Throughput: At peak cadence, we were running 137 consultants per quarter through the program across EMEA, with a consistent pass rate above 75%. The four-specialist dependency was gone within six months.
  • Delivery quality: CSAT scores on serverless engagements stabilized in the mid-9s and did not degrade as delivery volume increased. The judgment transfer worked.
  • Specialist leverage: The four original specialists went from delivering engagements to reviewing architectures and facilitating bootcamps. Their leverage increased rather than their workload. One of them later told me this was the first quarter in two years she was not behind on something.

What did not

  • The remote preparation was too unstructured in the first two cohorts. We gave participants access to lab environments and expected them to self-direct. About a third showed up underprepared. We added a structured lab sequence and the problem went away, but it cost us two cohorts’ worth of sub-optimal outcomes before we diagnosed it.
  • We tried to run the program with one facilitator per cohort to reduce cost. The residential phase requires two: one to run the room and one to observe the discussions and catch the misconceptions that do not surface in formal Q&A. Single-facilitator cohorts produced graduates with less confident judgment.
  • The pass/fail assessment generated anxiety that affected performance. Several consultants who were demonstrably competent during the case discussions performed below their demonstrated level on the formal assessment. We shifted to a portfolio model for the retake path, which improved both outcomes and participant experience.

What I would do differently

I would involve the four specialists earlier in curriculum design. I brought them in as facilitators but built the curriculum without them, because I did not want the curriculum to reflect their individual styles. In practice, the best curriculum material came from conversations with them about the failure patterns they saw repeatedly, and most of that happened after the program launched. Starting those conversations in month one would have improved the first two cohorts substantially.

I would also be more deliberate about what happens after the bootcamp. We invested heavily in getting consultants to working competency and lightly in keeping them there. Competency in a fast-moving technical area decays. A quarterly refresh mechanism (a case study, a new lab, anything that keeps the judgment active) would have maintained the investment better than we did.

Outcome

The program ran for three years. By the time the initial engagement closed, EMEA Professional Services had over 400 consultants who had cleared the serverless competency standard. The waiting list for serverless engagements was eliminated. The four original specialists led a team of practice leads rather than covering individual accounts.

The program structure was adapted by two other AWS regions. The case library that accumulated over three years of cohorts became a reference for the practice globally, which was not an original objective but turned out to be its most durable artifact.