What FAANG execs read this week: July 19-26, 2026

This week's strongest reading signal is a single open-models letter shared by Microsoft Chairman and CEO Satya Nadella and independently praised by former Google product director Manish Sainani. Ethan Evans adds one practical management article from Amazon Web Services Principal Engineer Anusha Dasarakothapalli. The public record is narrow, but both entries have enough first-party text to explain what the material argues and when it is useful.

Read this as a digest of visible public recommendations and article shares, not a complete map of private executive reading.

Satya Nadella shared Open Weights and American AI Leadership on July 24, writing that open-weight models are "essential to a healthy AI ecosystem" and can strengthen American competitiveness and economic opportunity while protecting national security. The linked page dates the letter July 24, 2026. 1 2

The document defines open-weight models as models that people can download, inspect, modify, and run on their own infrastructure. Its argument is practical: organizations should be able to match model size and cost to the job, rather than send every task to a frontier model or accept permanent dependence on one provider. It also argues that open models can improve competition, let customers retain more control over data and adapted systems, and give defenders more access to models for security testing. 2

The letter does not pretend that openness removes risk. It says released weights are difficult to trace or reverse, then rejects a blanket prohibition in favor of safeguards, evaluation, red teaming, and targeted legal or commercial remedies. It also calls for broader access to compute and shared training assets such as datasets, tools, and evaluation frameworks. 2

Manish Sainani called the document a "Great paper" and said he "thoroughly enjoyed reading" it. He turned the letter into three actions for engineers: run an open-weight model locally, move one real task from a frontier API and publish the cost, speed, and quality comparison, then ask what happens to data and fine-tuning if an AI vendor's price triples. 5

Sainani's current public LinkedIn profile lists Hushh Technologies, while a Google Cloud post identifies him as a former Director of Product Management for ML Infrastructure. That background makes his reading useful for managers who need to translate model policy into infrastructure, procurement, and product choices. 6 7

The document's own label is a letter; Sainani calls it a paper. That distinction matters less than the convergence: Nadella points readers to the policy argument, and Sainani supplies an engineering manager's test for whether the argument changes anything in practice.

On July 23, former Amazon VP Ethan Evans shared a guest article from Level Up and told readers to "read her piece here." The article is by Anusha Dasarakothapalli, a Principal Software Engineer in AWS's Applied AI Solutions Organization. The post describes her work across distributed systems, large-scale data platforms, and agentic AI infrastructure, where she leads initiatives across teams that do not report to her. 3 4 8

The article's central idea is the "Authority Trap": waiting for a manager or director title before trying to lead. Dasarakothapalli argues that influence comes from understanding other teams' constraints, reducing risk, and making progress visible before a formal decision is made. Her three-step playbook is to earn credibility with the teams involved, replace authority with tangible value, and build alignment before the meeting where the decision is supposed to happen. 4

The most concrete example is an AWS initiative tied to new regions across an organization spanning seven directors or general managers and more than 450 engineers. A request for a 5% headcount allocation stalled because teams saw it as an outside tax on already committed roadmaps. Dasarakothapalli changed course by talking with engineers about recurring operational pain, finding fixes they cared about, and returning to leadership with a plan that had already been tested and supported by the people who owned the systems. 4

The AI angle is specific rather than grand. The article argues that a prototype, small fix, or experiment can make an idea easier to evaluate, and that AI reduces the time needed to investigate a system and demonstrate a working improvement. The article is paid; the public version lays out the premise, the example, and the first two steps of the playbook, while the final step is behind the subscription wall. 4

Start with Open Weights and American AI Leadership if you are choosing models, reviewing an AI vendor, or deciding how much infrastructure your team should control. Read Sainani's three actions alongside it, then turn one of them into a small experiment with a measurable baseline.

Choose Dasarakothapalli's article if your current problem is cross-functional: a roadmap change that needs another team's support, a platform proposal without formal ownership, or a promotion case that depends on evidence of leadership before the title arrives.

The overlap between the two picks is useful. One asks who should control the underlying AI capability. The other asks how to move an organization when control is distributed. Both put the reader in the position of testing an idea through a concrete action rather than adopting it as a slogan.