hebel. today · 9 November 202631 / 74

Career · Product Management Evolution

Product teams shift from token-heavy experiments to evidence-based workflows

Validating product judgment. At one firm, a Gmail feature passed fake-door and usability checks but failed an A/B test after full implementation. Itamar Gilad reports that teams frequently ship unvalidated ideas, missing the resulting…

Validating product judgment. At one firm, a Gmail feature passed fake-door and usability checks but failed an A/B test after full implementation (Alex Ponomarev from Thriving In Engineering). Itamar Gilad reports that teams frequently ship unvalidated ideas, missing the resulting waste because they rarely measure launch outcomes against feature usage power laws. Repsol’s legal team compared Harvey against Microsoft Copilot and an internal prototype, Lexia, in a blind evaluation involving 50 lawyers across 12 jurisdictions (a16z). Harvey delivered the highest-quality output and saved lawyers four to six hours per week, leading to a 96% adoption rate across the department.

Automating cross-functional oversight. At one organization, agents use a refinement thread on Slack to ask questions from design, tech, and legal perspectives before development begins (Ops Forward). This workflow replaces weeks of handoff calls by logging decisions in a rules document and a Jira ticket as the machine ensures five perspectives appear in every decision. NVIDIA’s evaluation of specialized skills for its products raised average correctness scores from 46 to 87 out of 100 on specialized tasks (Salvador Carranza from What AI makes PossibNow). PayPal maintains its own skills through a shared registry that handles discovery, validation, and review, requiring updated skills to return through review before wider deployment.

Try today · 30 min

Create a versioned skill registry for your team's most frequent AI-assisted tasks.

Start by identifying one recurring workflow and assigning a single owner to track its current version and corrections.

You need

  • A shared registry space
  • A version control log
  • One identified recurring task

Steps

  1. Define the specific workflow for the agent.
  2. Assign one team member as the owner.
  3. Create a document to store the current version of the instructions.
  4. Add a section for collecting corrections from recent agent runs.
  5. Establish a review process for updating the instructions.

One email instead of your newsletter pile.

Forward the newsletters you already get (a Gmail filter does it) and tell Hebel your goals. Every morning it reads them with 20,000 others and sends one short email, with one small thing to do today. Fewer emails, not more. Free during beta.

Start my paper →
LinkedIn X

Sources

  1. a16z on hebel. →Analyzes how AI applications earn premiums by mapping pricing to recognizable work like research briefs rather than raw token consumption.

  2. Salvador Carranza, What AI makes PossibNow on hebel. → ✎ Written by a personDetails how NVIDIA and PayPal use specialized skills and registries to improve agent correctness and manage shared methods.

  3. Alex Ponomarev, Thriving In Engineering on hebel. → ✎ Written by a personExamines infrastructure decisions including logging and model choice to control token costs.

  4. Ops Forward on hebel. →Describes how growth design teams use Slack-based refinement threads to automate cross-functional input and maintain decision records.

More in Career

The conflict between task batching and deep focus

Identifying systemic incentives to improve team output

Iterate your career like a product launch

Turn daily friction into professional bonds

Demonstrate your value with concrete results

Product Management Evolution yesterday

Product teams debate the value of friction in AI-driven workflows

Sign in to Hebel first

Likes, your collection and done Hebel need an account. We email you a six digit code; no password.