GitHub plans capacity expansion and finer-grained write coordination

Independent

The Register reports four million additional CPU cores and 160 petabytes of storage on GitHub's roadmap following traffic growth and outages. GitHub describes changing write coordination from whole repositories to individual references, reporting three to four times higher throughput for writes to different references in internal tests.

Why it matters: GitHub's existing repository-wide lock makes unrelated branch writes wait through cross-region network delays. The planned coordination lets writes to different references proceed concurrently, addressing contention for developers and agents sharing busy repositories. Each write still incurs cross-region latency.

Decision: If expanding parallel agent work, do not treat the capacity roadmap or internal throughput results as evidence that production availability has improved.

What remains uncertain: The capacity additions are commitments, and the throughput measurements are GitHub's own. They do not establish improved platform availability. Writes that contend on the same reference still serialize.

Read: Reported article: GitHub powering up four million extra cores to stop its service going down; Primary source: The bridge to agent-scale Git infrastructure - The GitHub Blog; Primary source: GitHub Universe 2026: Evolving every layer of the stack - The GitHub Blog

Amazon retains AI usage metrics but warns against making them targets

Independent

Amazon published a memo explaining updated engineering role guidance centred on customer, business and team outcomes. It warns that targeting token consumption or code volume can increase costs and complexity, while retaining careful use of adoption and throughput metrics.

Why it matters: For engineering managers, the distinction affects performance assessment: Amazon recommends investigating whether heavy AI use produces matching delivery gains and why some teams are not adopting effectively. It also recognises documented cases where current tools do not help, allowing task-specific judgement within its push for adoption.

Decision: If AI usage informs performance reviews, use it to investigate delivery results and task suitability rather than reward consumption itself.

What remains uncertain: This is management guidance, not a demonstrated reduction in costs or complexity. Amazon claims usage metrics correlate with useful outputs but supplies no quantitative analysis in the memo. It continues to encourage engineers to learn and apply AI.

Read: Reported article: Amazon tells tokenmaxxing devs to slow their AI roll; Primary source: An update from Amazon engineering leadership on the role of the developer in the age of AI; Primary source: AWS CEO Matt Garman on why enterprises are seeing AI ROI

Telstra deploys an agent for billing reconciliation during migrations

Independent

Microsoft says Telstra's Migration Payment Agent supports more than 272,000 service migrations, with recommendation accuracy reported as 95%+. It identifies inconsistencies across legacy and new platforms, recommends corrections and prepares cases for human review.

Why it matters: For billing migration teams, the agent brings cross-system investigation and corrective recommendations into a workflow that includes human validation. This could reduce investigation effort, but the reported recommendation accuracy alone does not establish fewer billing errors or lower migration costs.

Decision: If considering a similar investment, require evidence of reduced reconciliation effort or improved billing outcomes before treating recommendation accuracy as proof of business value.

What remains uncertain: The scale and accuracy figures are Microsoft's claims, repeated by iTnews, rather than independent confirmation. The supplied documents do not define the accuracy measure or quantify review workload, financial savings or reductions in customer billing errors.

Read: Reported article: Telstra deploys billing reconciliation AI agent - iTnews; Primary source: The platform approach to becoming Frontier - The Official Microsoft Blog