The work behind an AI productivity claim

Proton addresses engineering bottlenecks, Ryanair proposes AI productivity measures, and new reporting examines who controls agent access to the web.

3 stories · Yesterday in AI

TL;DR

The short version
  1. Proton’s engineering memo calls for shared workflow investment; a faster draft still needs a path through testing, decisions and release.
  2. Ryanair includes AI in a wider cost review, with no separate savings target or measured return for the proposed work.
  3. Browser agents and dedicated connectors create distinct access routes; product teams need to define permitted actions and human control across both.

Big Moves

1 story

Ryanair puts AI productivity inside a wider cost review

Ryanair’s half-year statement includes greater AI use among proposed responses to higher fuel costs, without assigning a savings target to the AI work.

As a leader: Require a named workflow and a quality-adjusted savings measure before booking an AI productivity benefit.

As a practitioner: Document the handoffs and exceptions a proposed automation changes, including the person who resolves a failed case.

Ryanair Holdings plc ↗Vendor documentation

Enterprise Ops

2 stories

Proton’s AI manifesto puts the bottleneck beyond writing code

Proton’s newly published engineering memo calls for shared workflow investment; it describes an organizational direction rather than a measured productivity gain.

As a leader: Fund the shared bottleneck a delivery trace reveals, and keep accepted changes, waiting time and repair effort visible.

As a practitioner: Turn a repeated agent failure into a reusable correction, then check its effect on the next comparable task.

Proton / Bart Butler ↗Vendor documentation

The agentic web makes access a product decision

New MIT Technology Review reporting examines browser agents and dedicated connectors; earlier primary research explains why each route needs explicit control over actions.

As a leader: Decide which service actions agents may take and how a person can intervene across every supported access route.

As a practitioner: Evaluate a representative task through each supported route and inspect the resulting state and approval behavior.

MIT Technology Review / Grace Huckins ↗Reporting