acmqueue.bsky.social

AI Coding Benchmarks Need Composition Cards: Why the same model scores 70 percent and 6 percent, and what every benchmark should ship The audit is cheap for public benchmarks and impossible for hidden ones without disclosure. queue.acm.org/doi/10.1145/...

How Does Generative AI Change the Algorithmic Fairness Problem? In AI fairness, one size never fits all. With generative AI, one size may fit none, demanding fundamentally new approaches to evaluate and mitigate bias. queue.acm.org/doi/10.1145/...

Failure Is a Feature: Engineering Fault-Tolerant AI Agents Treating LLM agents as distributed systems and borrowing 40 years of reliability engineering to make them work Reliability, not capability, is the binding constraint on agents in production. queue.acm.org/doi/10.1145/...

The Coding Philosopher Out of My Other Mind What would it take to convince us that an AI is conscious? queue.acm.org/doi/10.1145/...

Running an Agentic Threat Model Workflow, abuse cases, and runtime controls Helping security and engineering teams extend established threat-modeling practices to autonomous systems without relying on the AI model to enforce critical security boundaries queue.acm.org/doi/10.1145/...

A Valid Tool Call Can Still Be an Invalid Effect Why consequential agent actions need a semantic read set AI agents expose an interface failure: A system can synthesize a transaction at runtime and then erase its read set before emitting the write. queue.acm.org/doi/10.1145/...

CAFE(S): Your Agent Is Only As Good As Its Context Even the latest models degrade when operating on unclear inputs. We present CAFE(S), a framework that evaluates context across five dimensions: clarity, actionability, fidelity, efficiency, and security queue.acm.org/doi/10.1145/...

Finding the Real Bottleneck in Large-Scale ML Training Real efficiency requires fixing the underlying systems In production training clusters, actual model execution frequently consumes only half of an accelerator’s theoretical capacity queue.acm.org/doi/10.1145/...

Signs of a Brewing Crisis Crises have the destabilizing power to shift incentives and invert power structures Identifying these circumstances allows operators and engineers to anticipate opportunities when adaptive resilience will be required queue.acm.org/doi/10.1145/...

What’s in Your Software Repair Kit? The things you wish you had before a crisis You can’t prevent or predict the next systems meltdown. But you can build resilience into your systems and organization before the next crisis strikes. queue.acm.org/doi/10.1145/...

Data Architectures in the Age of AI Business semantics must be machine readable and continuously synchronized Federation must become the default for operational workflows Governance must shift from controlling access to governing autonomous behavior queue.acm.org/doi/10.1145/...

How CHERIoT Provides Strong and Able Isolation Without an MMU Co-design of hardware and software offers stronger security and ease of use CHERIoT is optimized for small devices but showcases what is possible at any scale if you start by assuming CHERI. queue.acm.org/doi/10.1145/...

Kode Vicious When You Wish Upon a Prompt Relying on an LLM doesn’t teach how to write code Today’s students must be taught not how to spot an errant asterisk, but how all the components of a system work together. queue.acm.org/doi/10.1145/...

Arithmetic Without Numbers What happens inside an LLM when it tries to calculate with nothing but matrices How does a large language model do arithmetic when all it has are matrices—no fingers, no scratch paper, no columns of digits? queue.acm.org/detail.cfm?i...

Too Many DevEx Metrics, Too Little Guidance DevEx Metrics Compass: making sense of developer experience measurement As AI-augmented development becomes standard practice, engineering leaders face mounting pressure to demonstrate its impact. queue.acm.org/detail.cfm?i...

The Challenge of Efficiency Versus Isolation: Distributed systems lessons from serving inference The mechanisms that make production inference affordable are the same ones that dissolve the isolation and recovery guarantees built over two decades. queue.acm.org/detail.cfm?i...

Operations and Life Unethical Ways to Manage Technical Debt You can ignore it, or you can adopt more above-board techniques that reframe projects so that confronting tech debt is an upfront part of the plan. queue.acm.org/detail.cfm?i...

Where to Draw the Line Which decisions should not be handed over to LLMs? AI automation is only replacing the visible artifact of software engineering, not the actual job. We are demanding senior judgment while eliminating the work that produces it. queue.acm.org/detail.cfm?i...

The Coding Philosopher - Indirected Reality: Language, truth, and LLMs How can LLMs appear to know facts about the world when they are trained only on patterns in language rather than direct experience of reality? queue.acm.org/detail.cfm?i...

Where Does the Foundation Come From? AI is taking over the work that used to develop professional judgment. For engineers with grounding, AI is a multiplier; without it, AI creates a productivity illusion in which output outpaces understanding. queue.acm.org/detail.cfm?i...

Titan Transients and LLM Scalability The ghost in the machine Is the compute-efficient frontier (CEF) a universal constraint on the scalability of GenAI systems? It is a consequence of larger LLMs detecting inter-corpora correlations on all size scales queue.acm.org/detail.cfm?i...

Beyond Zero: Enterprise Security for the AI Era Beyond Zero is a new security paradigm designed for the AI era. Google’s vision for the future of this access model as well as a call for industry collaboration and standards development are outlined here. queue.acm.org/detail.cfm?i...

Goodbye, and thanks for all the Bikesheds! The only bit left of FOSS as we know it will be that users will be able to read the source code. As reproducible builds catch on, they might even be able to compile that, as long as they make no modifications. queue.acm.org/detail.cfm?i...

Kode Vicious: Pebbles on the Beach Keeping hold of the big questions There are fewer things more maddening than watching someone wander around a space bumping into ideas as if they’re in a pinball machine. queue.acm.org/detail.cfm?i...

EngThrive: Make It Fast and Easy to Do Great Work A durable model deployed at Microsoft that measures outcome-oriented engineering organizing productivity into three dimensions: speed, ease, and quality, with "thriving" as a well-being guardrail queue.acm.org/detail.cfm?i...

You Don’t Know Jack About Formal Verification: What can you confidently guarantee about your software? By eliminating the cost of writing proofs, AI is removing the biggest barrier to formal verification. queue.acm.org/detail.cfm?i...

In Code They Think; In Proof We Trust AI agents can be hijacked by prompt injection to exfiltrate sensitive data. Solution: Constrain the agent to express its plan as a Kotlin script, then statically verify the script against a security policy before execution queue.acm.org/detail.cfm?i...

The SPACE of AI: Real-world lessons of AI’s impact on developers AI is augmenting developers rather than replacing them. Effective integration depends as much on team culture and support structures as on the tools themselves. queue.acm.org/detail.cfm?i...

From Technical Debt to Cognitive and Intent Debt: Rethinking software health in the age of AI Cognitive debt is the erosion of shared understanding, when no one can explain how a system works. Intent debt is the absence of clear goals, constraints, and rationale. queue.acm.org/detail.cfm?i...

Eight Myths on Software Engineering and GenAI Examining the most common misconceptions This article provides practitioners, team leads, and engineering leaders research so the decisions organizations make about AI are grounded in evidence, not just enthusiasm. queue.acm.org/detail.cfm?i...

Climbing the Generative AI Mountain A “hitchhiker’s guide” for product managers Generative AI has arrived in the workplace, with high expectations and little guidance. PMs are expected to be more productive overnight, without a map for how to get there. queue.acm.org/detail.cfm?i...

The AI-Native Developer Redefining work, identity, and the future of craft In the context of pervasive AI, what parts of your job are still worth doing? Through surveys and interviews with practitioners, we investigate what it means to be a software developer today queue.acm.org/detail.cfm?i...

Knowledge Graphs over Two Decades: From web-scale extraction to LLM-augmented intelligence Xia Luna Dong The boundary between symbolic and neural knowledge continues to blur, leading to a new era of flexible, context-aware knowledge systems. queue.acm.org/detail.cfm?i...

Operations & Life The Important Decisions Document Every project (& family) needs one Long-running projects are a journey. An IDD helps new team members get up to speed, prevents wasting time on relitigating old decisions, improves morale, and increases accountability queue.acm.org/detail.cfm?i...

Kode Vicious: KV the Apostate Faith-based computing versus the unnatural science Whether we ask an LLM or a recent graduate to type the code is less important than knowing what the code does, how it was built, and when to look under the hood queue.acm.org/detail.cfm?i...

Bridging the Moat: Security for the Layperson Usability is core to effective security controls Cybersecurity is not just a problem facing enterprises. Individual users face similar risks without expertise and support to back them up. queue.acm.org/detail.cfm?i...

The Second-System Pit of Failure Lessons learned from building a second-generation system Our experience with building a new LMS to replace an existing system provides at least one point of evidence that it is possible to avoid the SSE trap. queue.acm.org/detail.cfm?i...

Building Malleable Systems, not Future-Proof Ones: Design for Change The code you write can’t possibly predict every change that comes along. Code is like a house, and people have to live in it, often for much longer than you expect. queue.acm.org/detail.cfm?i...

Kode Vicious Escape Routes Design your APIs carefully #Software is almost infinitely fungible, but at some point you have to make choices and pin down what you want a thing to do. This is the art of #API design. queue.acm.org/detail.cfm?i...

Open Source and the Iceberg Theory: Why “dependency management” isn’t enough anymore In an era of AI-generated code and increasing supply-chain attacks, stewardship is now a fiduciary and societal imperative. queue.acm.org/detail.cfm?i... @auggie.dev

Paul Vixie on the Evolution of Program State AI coding assistants mean more software will be created than ever before and by more people and agents than ever before. It's undecided as yet whether this era will be safer or more dangerous than the last. queue.acm.org/detail.cfm?i...

Data Analysis: Why Is It So Complicated? A sense of the depth and breadth of why it's so complicated to conduct and interpret data analysis. Why your models are incomplete and rife with inaccuracies, assumptions, caveats, and limitations queue.acm.org/detail.cfm?i...

Minimalist Design for Space Camera Flight Software While many systems drift toward more complexity, we advocate for a minimalist approach, with examples that have performed well in practice, with applications for other embedded software. queue.acm.org/detail.cfm?i...

Running the “Reflections on Trusting Trust” Compiler Revisiting Ken Thompson’s sourceless backdoor This article revisits that backdoored compiler, presenting the original code Thompson wrote more than 50 years ago. queue.acm.org/detail.cfm?i...

Kode Vicious: Driven to Distraction From floats to characters and back again 50 years of doing things the Unix way has left us bereft of better APIs. It's not just the drivers but also the application APIs on top of the operating system. queue.acm.org/detail.cfm?i...

Operations and Life: #SRE Is Anti-Transactional An API for interfacing with automaters Systems built by SREs are not fully autonomous on day one. It's iteration over time that leads to fully autonomous, functional, reliable service. queue.acm.org/detail.cfm?i... @yesthattom.bsky.social

A Practical Guide to Transitioning to Memory-Safe Languages Turning off the spigot of vulnerabilities: a new strategy for memory safety Traditional approaches to memory safety are flawed, treating symptoms without addressing the underlying cause. queue.acm.org/detail.cfm?i... @ayper.bsky.social

Practical #Security in Production Hardening the C++ Standard Library at massive scale Our collective experience at Apple and Google demonstrates that significant safety gains are achievable with surprisingly minimal performance overhead in production environments queue.acm.org/detail.cfm?i...

Safe Coding Rigorous modular reasoning about software safety Safe coding embodies a modular, compositional approach to building and reasoning about the safety of large, complex systems. queue.acm.org/detail.cfm?i...

Memory Safety for Skeptics If you're tired of hearing about memory safety, this article is for you. The state of possibility with memory safety today is similar to the state of automobile safety just prior to the widespread adoption of mandatory seat-belt laws. queue.acm.org/detail.cfm?i...