TL;DR Wrapping coding agent harnesses in a repeated plan-develop-QA loop boosts multi-day autonomous software development performance by an average of 52%. Title: Harness-of-Harness: Multi-Day Autonomous Software Development with Continual Improvement URL: Points 🔁 It leaves existing harnesses like Codex CLI and OpenCode untouched, adding an iterative improvement layer on top 🧭 Splits work into Planner, Developer, and QA Tester roles, carrying verified evidence forward into each new loop 📊 Improved across all 3 benchmarks and 3 model configs, e.g. GameCraft-Bench score jumping from 49.58 to 71.52 💰 Beats plain repeated Vanilla passes even at the same pass count, and is more token-efficient too 🎮 A 70-iteration, multi-day case study built a fully playable FPS game from scratch 🧩 Ablations confirm every piece matters: dropping plan updates, evidence feedback, or warm-start each hurts performance What stands out to me is the philosophy: not just running longer, but iterating smartly by accumulating verified knowledge. #AIAgents# #SoftwareEngineering#

"How is DDD actually used in open source, and how does it relate to maintenance quality?" — famously discussed, yet almost no large-scale evidence existed. This study tackles it by mining 865 repositories 🔍 Title: Domain-Driven Design in Practice: A Mining Study of Maintenance and Evolution in Open-Source Repositories URL: ❓ What does it study? 💡 Starting from 1,260 DDD-tagged GitHub repos filtered down to 865 (Java/C#/TypeScript), it analyzes eight tactical building blocks: Entities, Value Objects, Aggregates, Repositories, Domain Services, Domain Events, Application Services, Factories. ❓ How are they detected? 💡 A three-layer pipeline: DDD-specific annotations, naming conventions (e.g. OrderRepository), and package/directory structure — proceeding only when precision ≥0.75 is met. ❓ What's the biggest technical challenge? 💡 Bounded Context boundary violations. It operationalizes the "model-code gap" as cross-context dependencies, with violation rate = cross-BC deps / total inter-class deps. Boundary inference is human-validated requiring Cohen's kappa ≥0.80. ❓ How is rigor ensured? 💡 Kruskal-Wallis + Dunn post-hoc for cross-language tests, Spearman's rho for correlations, controlling age/size/language/type/team-size as confounders, with a pre-registered "degradation plan" that narrows scope rather than invalidating results if detection misses precision thresholds. The methodological rigor stands out — a solid empirical foundation for discussing DDD maintenance and evolution. #DDD# #SoftwareEngineering#

The gap between "what software actually does" and "what users believe it does" — this paper tackles that root problem with one unifying measure: explicit meaning 📐 Title: Making Software Meaningful URL: 📐 Overview The paper advocates adopting "explicit meaning" as a unifying measure to improve software usability, modularity, and accountability. It describes software behavior as a shared representation grounded in domain-specific language, not technical jargon. ❓ Challenges Solved Software engineering has a fundamental gap. ・"What the software actually does" diverges from "what users believe it does" ・That divergence breeds poor usability, bugs, and distrust The thesis: quality improves when developers and users hold a single, explicit understanding of what the software does and why. 💡 Methodology & Proposed Approach It organizes domain-specific phenomena (individuals, actions, and resulting facts) into concepts as a shared vocabulary, shown through three applications. ・Usability design: align user and designer perspectives via shared meaning ・Modular code generation: use LLMs to map conceptual units of meaning to code units, boosting modularity and legibility ・Agent accountability: implement codes of conduct based on explicitly defined intended behaviors 🌍 Use Cases It applies to improving modularity and legibility in LLM-assisted development, and to AI governance by making an autonomous agent's intended behavior explicit and accountable. It argues for putting "meaning" at the center as humans and AI build software together. #SoftwareEngineering# #AIGovernance#

From "AI as autocomplete" to "AI as an agent" 🚀 The March GitHub Enterprise Roundup highlights the industry shift: moving from simple AI suggestions to governed, measurable, and auditable AI that can handle real-world engineering tasks. As GitHub’s Dave Burnison notes, the pace of innovation right now is "mind-blowing," but the goal remains consistent: "To work alongside you to make you a better, more productive developer—not replace you." 🤝 Key updates this month: Agentic Development: With the Copilot coding agent and GitHub Agentic Workflows entering technical preview, AI is shifting from suggestions to delegated execution. This allows teams to delegate tasks like planning, triage, and CI/CD reasoning while maintaining human approval and enterprise controls. 🤖 Measurable ROI: Telemetry tools, like the Copilot usage metrics API, allow engineering leaders to programmatically track pull request throughput and time-to-merge, providing data-driven evidence for AI impact. 📊 Governance at Scale: The "Required Reviewer" rule is now GA, and new organization-level dashboards for Code Quality help leaders enforce standards across thousands of repositories without manual oversight. ✅ Security Incident Response: New enterprise-wide credential management tools allow administrators to rapidly audit and revoke compromised credentials during an incident, significantly reducing "time-to-contain." 🛡️ 🔗 Read the full March roundup here: #GitHubEnterprise# #SoftwareEngineering# #AI# #DevOps# #TechLeadership#

Community Invite: GitHub Social Club @ Amsterdam We’re hosting a GitHub Social Club in Amsterdam on Monday, March 23, and we’d love to see you there. This is a low-key hangout designed for developers, builders, researchers, founders, and the open-source community to unplug and meet others who care about the craft of building software. The Vibe: No talks, no pitches: Just good conversations, coffee, and snacks. Meet the team: Connect with Hubbers building GitHub, Copilot, and MCP. Swag: We’ll have GitHub gear and plenty of stickers. Flexible: Come solo, bring friends, drop in for a bit, or stay awhile. Event Details: 📍 Location: Ceintuur Theater, Amsterdam 📅 Date: Monday, March 23 ⏰ Time: 10:00 AM – 4:00 PM RSVP or Drop-in: If you know developers or communities who would value this space, please feel free to share this post and invite them along. Hope to see you there. #GitHub# #KubeCon# #CloudNativeCon# #OpenSource# #SoftwareEngineering#

Software engineering is going to feel more like content production / journalism. Ability to gauge demand / vitality becomes more important, but it's ephemeral. Stunts, gimmicks, & culture become more important. Brands become more important.