Editorial note:
Software that used to "just work" is being redefined — through expiring certificates, through AI changing who builds systems, and through careful rewrites that trade features for safety. Today's pieces all trace a single theme: control — over your tools, your knowledge, and your infrastructure — is where power still lives.
In Brief
OpenBSD’s openrsync lands in base
Why this matters now: openrsync gives sysadmins a cleaner, security-minded rsync alternative that’s now part of OpenBSD base, and teams should assess compatibility and threat-model changes before switching.
OpenBSD upstreamed openrsync, a purposefully smaller, security-oriented reimplementation of rsync’s sender/receiver algorithm. The project emphasizes correctness and hardening: it leans on OpenBSD primitives like pledge(2) and unveil(2) to limit what processes can do, and intentionally supports a subset of rsync flags to reduce complexity.
"openrsync is about 10 000 lines of C code: do you trust me not to make mistakes?" — a candid line from the repo that underlines the trade-offs.
Key takeaway: openrsync is attractive if you want a tidy, auditable implementation and strong sandboxing on OpenBSD; but be cautious about protocol quirks (trailing-slash behavior, metadata) and the weaker security guarantees on non-OpenBSD ports. See the repo for details.
Accenture to acquire Ookla
Why this matters now: Accenture is buying Ookla largely for telemetry and benchmarking data, signaling consultancies are doubling down on data assets to power AI-driven network operations.
Accenture announced it will acquire Ookla, the company behind Speedtest and other measurement products, folding those telemetry streams into Accenture’s data and AI practice. Commenters noted the obvious: the front-end testing apps are easy to copy — the long-lived value is the user base, relationships with telcos, and the measurement dataset that operators pay for. Expect scrutiny over measurement neutrality and how private measurement feeds will be monetized. The press release frames this as an enterprise play to optimize Wi‑Fi and 5G networks at scale.
Voxel Space — a nostalgic lesson
Why this matters now: Voxel Space’s constraints-driven tricks are a useful reminder that smart algorithms can deliver impressive results even without modern GPUs — a useful perspective as people optimize for cost and latency.
A well-done retrospective on NovaLogic’s 1992 Voxel Space engine walks through how textured heightmaps and column projection gave long draw distances with tiny CPU budgets. The write-up is a neat primer for game devs and systems folks who want tricks that trade compute for precomputed data. Read the walkthrough and demo at the Voxel Space page here.
Deep Dive
Microsoft Office 2019 and 2021 for Mac view-only conversion
Why this matters now: Microsoft’s certificate expiry will push affected Office 2019/older Office 2021 macOS and some iOS installs into a "reduced functionality mode" on July 13, 2026 — meaning perpetual-license customers may lose editing on products they paid for.
Microsoft uses a signing certificate to validate certain Office installs on macOS and iOS. According to an explainer and community thread, that certificate expires on July 13, 2026, and affected copies of Office 2019 (which have no update path) and some older Office 2021/iOS builds will reportedly switch to a Microsoft-defined "reduced functionality mode." After the expiry, those apps "can open and view files but can't edit, save, or access full features." Microsoft quietly edited an older support page — removing a prior 2023 statement that the apps would "continue to function" — and is emailing affected customers with migration options: free web apps, a Microsoft 365 trial (which requests a payment method), or buying a subscription or new perpetual license.
This has provoked two kinds of responses. One line is practical: users who depend on local editing need a migration plan today — switch to another editor, install a maintained Office build, or move documents to a platform Microsoft continues to support. The other line is philosophical and legal: perpetual-license buyers feel like they've lost ownership. Hacker News threads are already raising consumer-rights and regulatory questions (for example, remedies under Australian law), and urging moves to alternatives like LibreOffice or Pages.
"Can open and view files but can't edit, save, or access full features." — the description circulating in notices to affected users.
What to do now: if you're on Office 2019/old 2021 on macOS or certain iOS builds, audit your installs, back up editable copies, and test alternatives. Don’t assume "perpetual" means immutable — platform dependencies like certs can create single points of failure. Vendors should also be judged on their communication: quietly changing a support page and then emailing limited-conversion options is likely to be litigated or at least provoke regulatory scrutiny.
Domain expertise has always been the real moat
Why this matters now: As agentic AI starts generating production code, people who can judge domain correctness — not just code correctness — will be the most valuable contributors in product and engineering teams.
The thesis from this thoughtful post is simple and consequential: historically, writing working software meant two hard things — translating domain knowledge into behavior and implementing it. Agentic AI makes the implementation easier; the scarce skill becomes judging whether outputs match reality. As the author puts it, "the binding constraint has moved from can you build it to can you tell whether it’s right."
That reframes hiring, roadmaps, and tooling. Teams that once prized generalist engineers who could both spec and ship might now prioritize hybrids: domain experts who can prompt and validate agents, and engineers who can manage correctness checks and safety nets. Commenters point to Polanyi’s paradox — many domain experts "know it when they see it" without being able to codify rules — and the post recognizes practical bridges like example-driven workflows, domain-specific languages, and behavior-driven development as ways to externalize tacit knowledge.
The practical implication is immediate. Product managers and engineering leads should invest in mechanisms that make domain knowledge explicit and testable: curated datasets for prompts, executable acceptance tests, and human-in-the-loop validation checkpoints. AI writes code fast; humans must teach it what matters. Teams that build those feedback loops will get the best returns from AI assistance.
Closing Thought
Perpetual licenses can fail because of external dependencies, models of development change what skills are scarce, and sometimes the best fix is to rewrite a tool smaller and safer. That’s a pattern: control shifts from the monolith to the boundary — certs, data, and expertise — and whoever controls the boundary shapes outcomes. Keep copies of your data, insist on clear vendor promises, and invest in domain thinking: those are practical hedges for a world where the tools are evolving faster than our mental models.