Key Takeaways
- Tw93 published a concrete GitHub Actions optimization playbook, arguing AI speeds up coding but CI verification remains the bottleneck.
- Arvid Kahl argued that human-in-the-loop review will become a burden as agent costs fall and agents stay fully focused without fatigue.
- Tw93 launched a Mole Pro beta with bandwidth, speed-test, and device-info views while keeping the core app simple.
- @levelsio flagged a proposed Italian exit tax on unrealized gains starting 2027, framing it as part of a wider European tax trend.
- Tony Dinh joked about a homegrown tool versus Cursor and described 13-hour days as an indie developer.
1. AI-Assisted Development and CI Bottlenecks
- Tw93 said AI has made writing code much faster, but verification is still slow, especially GitHub Actions CI. He optimized CI for Mole, Kaku, and Pake and shared an agent-ready prompt: pull the last 50 runs with
gh, split each job into queue, setup, and test phases to find bottlenecks; cancel superseded runs on the same PR withcancel-in-progresswhile lettingmainfinish every commit; and usepathsfilters for docs-only changes, noting that skipped required checks can block PR merges. He also recommended setup-action caches (setup-nodecache,setup-godefault,Swatinem/rust-cachefor Rust) within the 10 GB per-repo cache limit, checking that a second local build is near-instant to catch Dockerfile or script issues, downloading prebuilt tool binaries instead ofgo install/cargo install, parallelizing dependency-free jobs, removingneedsthat pass no data, and splitting the longest tests across machines. He added moving OS-agnostic jobs like formatting, lint, and dependency audits to Ubuntu, avoiding redundant checks such as pulling a full Rust cache for a frontend-only job, and measuring before/after timings after each change 1.
2. Agents and the Changing Shape of Work
- Arvid Kahl argued that "just tweak it manually" often has side effects the editor did not foresee, such as changing one padding value and breaking another area or a responsive layout, and that handing such seemingly harmless changes to an agent noticeably reduced the fixes needed. He endorsed a statement attributed to Proton's CTO on AI, saying its points matched what he had learned over roughly the past six months and that the industry is changing and practitioners must change with it. He said The Mythical Man-Month assumed high communication, documentation, and feedback costs, and that while agents still need these steps, their cost is falling daily and their speed rising, which he called possibly the biggest part of a paradigm shift in "work." He predicted human-in-the-loop will soon become a burden: a well-configured agent will not rubber-stamp work to save time, stays 100% focused every time, and does not tire, whereas humans are distracted even when focused, making agents clearly better for implementation and operational tooling. He also said customer-facing AI is hard to build 1 2 3 4 5.
3. Product Launches and Indie Developer Signals
- Tw93 announced Mole Pro, which keeps the app simple while adding a view of which app is using bandwidth, a speed test, and hover labels showing Wi-Fi, disk, and battery internals; it is available via Settings > Receive beta updates 1.
- Tony Dinh joked about a self-built or alternative tool with "we already have Cursor at home," implying a gap versus Cursor, and said he works 13 hours a day after quitting his 9-to-5, mocking indie life 1 2.
4. Policy and Market Signals
- @levelsio linked European tax proposals into a trend: after France's "Gigabyte tax" and Germany's "AI tax," Italy reportedly plans a 26% exit tax from 2027 on unrealized investment gains for people with relevant assets above β¬800,000 who were tax residents for at least 6 of the past 10 years. He read this as Europe lacking money and governments trying to seize assets 1. He also cited the claim that "software is dead, the personal software era is here" and asked what remains to be done once AI can build everything, answering that much remains 2.
