Editorial note: Today’s pick is about where defaults and tooling quietly reshape outcomes — who can access services, which communities keep control of data, and how much money your LLM deployments actually burn.
In Brief
I Don't Have a Smartphone
Why this matters now: Author Ploum's refusal to carry a smartphone highlights how phone‑and account‑first service flows are turning basic errands into exclusionary, surveillance‑laden gatekeeping.
The piece opens bluntly: “I don’t have a smartphone,” and uses that line to walk through the many silent assumptions modern services make — everyone has a charged, uncrowded phone; everyone can install apps; everyone trusts a Google/Apple account — and how each assumption can break real lives. The author details a practical workaround of juggling three devices (an e‑ink Mudita Kompakt, a /e/‑flashed FairPhone, and an old OnePlus) to avoid platform lock‑in and surveillance, while still being usable in a world that increasingly refuses to accept anything else.
“Those 10 assumptions have become the new commandments everyone should obey.”
This isn’t just a hobbyist posture. The post — and the lively discussion on Hacker News — surfaces concrete harms: parking systems that penalize people who don’t use an app, banks that hard‑require specific mobile stacks for MFA, and everyday social friction that turns non‑phone users into second‑class citizens.
SteamDB Joins Nexus Mods
Why this matters now: Pavel Djundik’s sale of SteamDB to Chosen (the company behind Nexus Mods) could change how a major community resource is funded, run, and monetized — with immediate implications for modders and data users.
SteamDB, a community‑built tracker for Steam data, has been acquired by Chosen/Nexus Mods. Djundik framed the move as a necessity to keep the service alive and growing; Nexus vows to invest resources and not “change what makes SteamDB special.” The announcement and thread quickly split the community: optimism about stability and integration with Nexus, but worry about the usual post‑acquisition hazards — ads, gated features, or shifts in governance.
“Our job isn't to change what makes SteamDB special. It's to give it the resources, team and long‑term stability it deserves,” wrote Nexus’ leadership.
Users are already talking about mirror plans and scraping historical backups “just in case,” which is a telling reaction when trust in long‑running community projects is fragile.
Making the Internet Boring
Why this matters now: Cemrehan Cavdar’s argument for introducing friction into the web offers concrete tactics to blunt algorithmic hooks — an immediate, low‑tech countermeasure anybody can try.
The essay suggests that deliberately dimming the internet’s shine — turning off recommendation features, clearing watch histories, or imposing small frictions like typing URLs — is an effective way to reclaim attention. Practical tips from the original post and thread range from whitelisting apps with device management tools to psychological strategies that reduce compulsion without outright banning discovery.
“When you fight something, you're tied to it forever,” a commenter paraphrased, arguing that softer decoupling sometimes works better than heavy‑handed controls.
If you feel the feed is running you, this is a short playbook with immediate, reversible interventions.
Deep Dive
FrontierHarness Eval – 9 harness, same model, cost per pass varies 17x
Why this matters now: FrontierHarness’s cross‑harness test shows that orchestration and tooling choices — not just model selection — can change your per‑task cost by an order of magnitude, which is critical for budgeting and procurement of agentic systems.
FrontierHarness ran one model across nine different agent harnesses for terminal‑style software engineering tasks and reported an eye‑watering result: median cost per successful pass varied by up to 17x depending on the harness used. The team kept hardware, dataset, and model checkpoint constant (runs executed on Runta with a golden checkpoint) so the harness was the variable. That design isolates where inefficiency lives: prompt handling, caching, tool calls, control flow, and runtime overheads.
“Harness wars are the next browsers wars!!” — a Hacker News reaction that sums up how folks immediately grasp the commercial stakes.
Why does that matter? As models become more fungible, the layer you wrap around them — your orchestrator, toolset, and prompt pipelines — becomes the decisive lever for cost, latency, and reliability. A harness that does smart caching, minimizes redundant token usage, or batches calls efficiently can save you thousands of dollars at scale. Conversely, a harness that naively calls the model for every micro‑step will multiply costs even if you picked the “best” model.
The report is deliberate about limits. FrontierHarness v1.0 focuses on software engineering contexts and terminal tasks; a few harnesses in the set are intentionally minimal, and some harnesses show model‑specific quirks that can skew results unless you add per‑model mitigations. The repo currently lacks plug‑and‑play instructions, so independent reproduction is nontrivial. Still, the practical takeaway is clear: procurement conversations should treat harness choices as first‑class items. Budget lines, SLOs, and vendor comparisons need to include orchestration metrics, not just model price per token.
What should engineers and managers do now? Start with narrow, reproducible microbenchmarks that match your workload, include harness overheads in cost estimates, and prefer harnesses with transparent instrumentation so you can trace where tokens and compute are spent. Expect more follow‑ups — FrontierHarness plans broader matrix tests (harness × model) — and treat today’s results as a strong signal, not a final verdict.
Closing Thought
Defaults and tooling quietly decide who gets served and who pays. Whether a vendor assumes you have a smartphone, a community worries what an acquisition will do to its data, or an agent harness multiplies costs, the non‑sexy engineering choices are the ones actually shaping outcomes. Notice them — and budget, design and govern accordingly.