Credentials
Press commentary, peer review, research data and standards work.
Cited in the press
Expert commentary in the technology press.
AI Breakthroughs in Digital Health: Game-Changers in Care Delivery
Expert commentary (MD, PhD): keep doctors in control of AI. AI goes first into operational work (product, analytics, technical support); an AI assistant for doctors helps consult patients but does not replace the doctor’s judgment; on vague tasks the output looks fine but is wrong, so a person checks the result
Vendor lock-in becomes technology’s hidden tax
Expert commentary: vendor lock-in gives technology providers excessive control over customers’ options; the protection belongs in the contract: data exports in open formats and advance notice of pricing or configuration changes
An AI Study Without Its Data Is Fiction
Bylined article (Health Tech): an AI study without its data is fiction; all 36 agent runs are open, the raw data let me find my own cost error and publish a correction; reviewers should reject papers that come without the data or without saying how it was collected and calculated
An Experimental Comparison of Six AI Agent Session Reset Policies
Bylined article: 36 agent runs, six context-reset policies — clearing after every task costs about a third more than the optimum, which sits at three tasks; data and DOI 10.5281/zenodo.22759217
Access control after the human
Bylined article: access control when the credential holder is a program — access per task, logs the agent cannot edit, a named owner for every automated rule
Experts Comment: Is Anthropic's IPO Warning About Existential Risk Legal Caution Or Strategy?
Expert commentary: risk factors in an IPO filing are a legal duty, not a plot; the real risk for startups is a heavy rule written for the biggest labs landing on small companies
The One Piece Of Cybersecurity Advice Every Business Should Follow? Experts Weigh In
Expert commentary: treat an AI agent like a new employee account — access for one task, logs the agent cannot edit, and logins to outside platforms done by a human
What Can Startups Learn From HSBC's AI-Driven Job Cuts?
Expert commentary, quoted in the body of the article: if a test can check the result, an agent gets the work; if trust is needed, a human does; an AI voice agent on calls did not stick
Nobody with a crawler token asked for my llms.txt - what I saw in 15 days of nginx logs
Bylined article: 15 days of nginx logs, 46,155 requests carrying search and AI crawler tokens, none for llms.txt; the first numbers were wrong and were corrected (report DOI 10.5281/zenodo.23018429)
The Rise of the Thought Leader Creator: What Brands Need to Know in 2026
Expert comment: in my own data (326 videos, nearly 5,000 posts) talking-head opinion did 24% worse and concrete take-away content 34% better; saves relative to views as the earliest signal
The Writing Got Cheap. The Checking Did Not
Bylined article (Health Tech): producing an analysis with AI got cheap, checking it did not, and the checking is what the work is for
97% of what my coding agent billed for was re-reading its own context
Invited reprint of the DEV.to article (canonical → arsentev.ai/research): what coding-agent sessions really pay for is re-reading context
My company has no full-time developers, DevOps or designers. Here's what three $200 Claude subscriptions do instead
Bylined column: businesses with no full-time developers, DevOps or designers — AI agents on three $200 Claude subscriptions instead of ~$38,500 a month at BLS median pay
All my agent's tests were green, and they told me nothing
Bylined article: 36 coding-agent runs, 4,086 tests, zero failures — and exactly why a clean sweep could not compare the conditions
What Do Dario Amodei And The JCHR's AI Warnings Mean For Startups and SMEs?
Expert commentary: whatever AI bill lands will ask for a record of how systems are used; companies that already keep it will pass in an afternoon
UK SMEs Have Increased AI Spend By 1,000% - But Is It Actually Paying Off?
Expert commentary: the rise in SME AI spend is a billing-model shift, not a preference shift — the audit that matters is what the tools pay to re-read, not how many tools you pay for
97% of what my coding agent billed for was re-reading its own context
Bylined article: 97.05% of billed tokens went on the agent re-reading its own context, not on generation
Experts Comment: Is Britain's Reliance On Foreign Technology Becoming A National Security Risk?
Expert commentary: technological sovereignty is a question of substitutability, not ownership
What smart locks inherit from software
Bylined article: the failure modes a connected lock inherits from software, and where the trust boundary now runs
Going Offline Doesn't Remove Password Risk. It Swaps It.
Expert commentary: offline versus cloud password managers — which failure mode you choose
Anthropic Discloses Fourth Unauthorised Claude Access Incident: Is The Security Industry Prepared For AI Breaches?
Expert commentary: AI agents with legitimate access and enterprise threat models
Peer review
Open peer review for Qeios and PREreview, each review with its own DOI.
- 2026-10-23·PREreview·Forged user turns surviving context compaction in coding agents: bounded sessions as another exit, where long sessions sit, independent checks
Compaction Laundering: Provenance Destruction in Context-Compacting Coding Agents
- 2026-10-23·Qeios·Diffusion heatmaps as an image channel for chest X-ray retrieval and classification: an ablation row repeats retrieval numbers, a table caption disagrees with the data
DAug: Diffusion-based Channel Augmentation for Radiology Image Retrieval and Classification
- 2026-10-22·Qeios·Language 3D maps for robots: a tie with OpenScene on real encoder input, a text-formula mismatch in the smoothing weight
- 2026-10-22·PREreview·Experience memory for coding agents: where the 17% saving comes from, the cache cost of eviction, validation as the trigger
REM: What to Store, When to Recall, What to Forget in Experience Memory for Coding Agents
- 2026-10-21·Qeios·Environmental and public health harms of AI across its lifecycle: a narrative review
Adverse Environmental and Public Health Effects of Artificial Intelligence: A Narrative Review
- 2026-10-20·Qeios·Auditing an LLM hallucination-detection benchmark, sixth-round review: held-out threshold check, table reconciliation
- 2026-10-20·Qeios·A metagenomic foundation model for wastewater pathogen monitoring: a non-learned baseline, seeds, matched tuning budgets, held-out data for anomaly detection
METAGENE-1: Metagenomic Foundation Model for Pandemic Monitoring
- 2026-10-20·PREreview·Accepting a coding agent's work against the PRD: green tests versus delivered requirements, who accepts the work, empty methods and the judge, repeated runs
- 2026-10-19·Qeios·Chatbot versus GP consultation acceptability
- 2026-10-18·Qeios·Causal modeling against gender bias in medical imaging
Fair Diagnosis: Leveraging Causal Modeling to Mitigate Medical Bias
- 2026-10-18·Qeios·Survey of misalignment in vision-language models
Large Vision-Language Model Alignment and Misalignment: A Survey Through the Lens of Explainability
- 2026-10-16·Qeios·Auditing an LLM hallucination-detection benchmark, fifth-round review
- 2026-10-13·Qeios·Self-audit of an LLM hallucination benchmark, fourth-round review
- 2026-10-13·Qeios·Channel knowledge maps via image super-resolution
Deep Learning-Based CKM Construction with Image Super-Resolution
- 2026-10-15·Qeios·Shortcut learning in LLMs on natural language inference
Do LLMs Overcome Shortcut Learning? An Evaluation of Shortcut Challenges in Large Language Models
- 2026-10-11·Qeios·CNNs with fixed pre-defined edge filters
Convolutional Neural Networks Do Work with Pre-Defined Filters
- 2026-10-10·Qeios·Classifying workforce-intelligence AI under the EU AI Act
Decision Support or High-Risk? Classifying AI Workforce-Intelligence Systems Under the EU AI Act
- 2026-10-10·Qeios·AI reviewers vs humans across three disciplines
- 2026-10-08·Qeios·LLM benchmark for formal verification proofs
- 2026-10-08·Qeios·Adaptive ultrasound scan-line subsampling with a normalizing flow
Deep Sylvester Posterior Inference for Adaptive Compressed Sensing in Ultrasound Imaging
- 2026-10-09·Qeios·Survey of deep learning in medical image registration
Deep Learning in Medical Image Registration: Introduction and Survey
- 2026-10-07·Qeios·Public connections versus formal records in digital health startups
- 2026-10-07·Qeios·Rule-based LLM hallucination detector, third-round review
- 2026-10-06·Qeios·Post-training pruning of Stable Diffusion components
Efficient Pruning of Text-to-Image Models: Insights from Pruning Stable Diffusion
- 2026-10-06·Qeios·Hierarchical reinforcement learning with landmarks for Sokoban
Solving Sokoban using Hierarchical Reinforcement Learning with Landmarks
- 2026-10-06·Qeios·Bayesian-optimized hard negatives for vision-language navigation
Fine-Grained Alignment in Vision-and-Language Navigation through Bayesian Optimization
- 2026-10-06·Qeios·Deep reinforcement learning for protein folding in the 3D HP lattice
Protein Structure Prediction in the 3D HP Model Using Deep Reinforcement Learning
- 2026-10-05·Qeios·Influence functions for training data attribution in diffusion models
Influence Functions for Scalable Data Attribution in Diffusion Models
- 2026-10-05·Qeios·Dataset distillation via trajectory matching
Enhancing Dataset Distillation via Label Inconsistency Elimination and Learning Pattern Refinement
- 2026-10-05·Qeios·Parameter-efficient multimodal adapters for video transformers
CM3T: Framework for Efficient Multimodal Learning for Inhomogeneous Interaction Datasets
- 2026-10-05·Qeios·Noise level correction for diffusion model sampling
Enhancing Sample Generation of Diffusion Models using Noise Level Correction
- 2026-10-04·Qeios·Adapting video diffusion models into action-conditioned world models
- 2026-10-04·Qeios·Multi-language code vulnerability detection with CodeBERT
MVD: A Multi-Lingual Software Vulnerability Detection Framework
- 2026-10-04·Qeios·Partial convolution with attention for efficient backbones
- 2026-10-04·Qeios·Vision-language model semantic tokens for person re-identification
When Large Vision-Language Models Meet Person Re-Identification
- 2026-10-02·Qeios·LLM-based tabular data synthesis for recommender systems
SampleLLM: Optimizing Tabular Data Synthesis in Recommendations
- 2026-10-02·Qeios·GenAI evidence pollution in out-of-context misinformation detection
Mitigating GenAI-powered Evidence Pollution for Out-of-Context Multimodal Misinformation Detection
- 2026-10-02·Qeios·Efficient self-supervised speech pre-training with Zipformer
k2SSL: A Faster and Better Framework for Self-Supervised Speech Representation Learning
- 2026-10-02·Qeios·Training-free token merging to accelerate vision-language models
iLLaVA: An Image is Worth Fewer Than 1/3 Input Tokens in Large Multimodal Models
- 2026-10-01·Qeios·Bayesian inverse RL for LLM alignment
Approximated Variational Bayesian Inverse Reinforcement Learning for Large Language Model Alignment
- 2026-10-01·Qeios·Multimodal LLM for video moment retrieval
LLaVA-MR: Large Language-and-Vision Assistant for Video Moment Retrieval
- 2026-10-01·Qeios·An LLM game engine driven by plain-text rule instructions, poker case
- 2026-10-01·Qeios·A Chinese counterspeech dataset and LLM-as-a-judge
- 2026-09-29·Qeios·Temporally consistent planning in end-to-end driving
Don’t Shake the Wheel: Momentum-Aware Planning in End-to-End Autonomous Driving
- 2026-09-30·Qeios·Weight superposition in transformers as mixture of experts
Superposition in Transformers: A Novel Way of Building Mixture of Experts
- 2026-09-30·Qeios·Reliability metrics for enterprise skills graphs
Can You Trust the Graph Beneath the Agent? A Reliability Metrology for Enterprise Skills Graphs
- 2026-09-30·Qeios·VLM benchmark for thermal, depth and X-ray reasoning
Are Vision-Language Models Truly Understanding Multi-vision Sensor?
- 2026-09-30·Qeios·Duplication and mislabeling in vulnerability repair datasets
- 2026-09-29·Qeios·Fine-tuning Nougat for Arabic OCR to Markdown
Arabic-Nougat: Fine-Tuning Vision Transformers for Arabic OCR and Markdown Extraction
- 2026-09-29·Qeios·Offset-specific projections unifying attention and convolution
Translution: Unifying Self-Attention and Convolution for Adaptive and Relative Modelling
- 2026-09-29·Qeios·Neural activations as distance, not intensity, metrics
- 2026-10-06·PREreview·What an agent harness really sets is the bill: cache writes as their own line, cost with and without compaction, whether the agent ran tests before it stopped, trajectories with a DOI
- 2026-10-07·PREreview·Keeping control of agent-written code: what the reviewing agent was briefed to do, whether the test suite can tell the compared conditions apart, how false-green signals were found, whether the rulebook kept growing
- 2026-10-08·PREreview·Parallel sub-agents on long coding tasks: what the extra tokens consist of, how large the main agent context gets, what validating sub-agents were briefed to do, whether a unique success can be told from noise with one run per cell
- 2026-10-09·PREreview·Where agent tokens go in vulnerability discovery: what a stage token share really means, the price of cutting history, runs that never tested a candidate, one run per cell
Where Do the Tokens Go? Understanding and Reducing Costs in LLM Agents for Vulnerability Discovery
- 2026-10-10·PREreview·An agent's always-loaded instruction file: paid once per session or on every call, deleting lines by value, on-demand skills, the task check as the gain, repeats and rule order
Curating Always-Loaded Context for LLM Agents: A Capacitated Assortment Model with Censored Feedback
- 2026-10-11·PREreview·When compacting an agent’s context hurts: the price of blocking compaction, first compactions, harm measured by next actions, open data
- 2026-10-12·PREreview·Agents build working code but weak tests: all-green proves little, checks that cannot tell versions apart, a second sceptic agent, open data
TestSuiteBench: Coding Agents Struggle to Generate Strong Test Suites from Specifications
- 2026-10-13·PREreview·Review agents read code but rarely run it; a second agent briefed to disprove; where the cost of long sessions goes; open data
(When) Should a Software Factory be Lights-off? Benchmarking Code Review Agents on Bug Detection
- 2026-10-16·PREreview·Agent cost is driven by turns, not observation size: the expensive tail of long sessions, cache in real bills, context clearing as a knob, which turns are redundant
TurnCost: Turns, Not Observations: Cache-Aware Loop Engineering for Coding Agents
- 2026-10-17·PREreview·Code quality and the cost of agentic code review: the thin tail of expensive PRs, tokens versus a cache-priced bill, paging through long files, the cost of a verified finding
- 2026-10-19·PREreview·What to fill an agent context with: whole files versus compression, context paid for on every call, a patch-applies check in the loop, repeats and the empty-patch check
Skip the Compressor: Coding Agents Fix More Issues When They Pick Whole Files
- 2026-10-18·PREreview·What to put in AGENTS.md: the file's cost in the sessions that read it, a short file with on-demand skills, a check the agent can run, repeated runs
The Role of Agentic Instruction Categories for Code Generation: A Study of Five Python Libraries
- 2026-10-15·PREreview·Context compaction in coding agents: where the bill sits, when to compact, confident failures without a check, open data
- 2026-10-05·PREreview·What a silent verification suite proves: report fields next to pass/fail, who wrote the checks, an outside refuter, the hash and the artifact
Discriminating Fixture Coverage in Agent-Infrastructure Verification Suites
- 2026-10-04·PREreview·A second model as reviewer: the review prompt and a disprove-it stance, false positives, who wrote the answer key, tokens and open data
When Does a Second Model Help? Cross-Model Review in LLM Verification
- 2026-10-02·PREreview·When a coding agent should compact its context: compaction count, cost with caching, checking summaries, run-to-run spread
AutoCompact: Learning When to Compact Context in Long-Horizon Coding Agents
- 2026-10-01·PREreview·Self-verification in terminal agents: own tests, an independent refuter, LLM-judge agreement, cost
Can Terminal Agents Trust Their Own Verification? Diagnosing and Improving Self-Verification
- 2026-09-30·PREreview·Agent context compression: cost with caching, how often to compress, independent checks, repeats
FOCUS: Training-Free Decision-Preserving Context Compression for LLM Agents
- 2026-09-29·PREreview·Forecasting agent token use: cache, the expensive tail, context compaction, budgets
TokenCast: Forecasting Token Consumption During LLM Agent Execution
- 2026-09-28·Qeios·Few-shot visual inspection with in-context learning
Vision-Language In-Context Learning Driven Few-Shot Visual Inspection Model
- 2026-09-28·Qeios·One-shot federated fine-tuning of foundation models
One Communication Round is All It Needs for Federated Fine-Tuning Foundation Models
- 2026-09-28·Qeios·Open-world 3D detection benchmark for autonomous driving
OpenAD: Open-World Autonomous Driving Benchmark for 3D Object Detection
- 2026-09-28·Qeios·Accelerating diffusion transformers via skip-branch caching
Accelerating Vision Diffusion Transformers with Skip Branches
- 2026-09-26·Qeios·LLM-guided content-adaptive image tokenization
- 2026-09-26·Qeios·Grounded image quality assessment with multimodal LLMs
Grounding-IQA: Multimodal Language Grounding Model for Image Quality Assessment
- 2026-09-26·Qeios·Equivalent transformer decomposition into mixture-of-experts
Union of Experts: Adapting Hierarchical Routing to Equivalently Decomposed Transformer
- 2026-09-26·Qeios·Parameter-efficient fine-tuning via LDU decomposition
LoLDU: Low-Rank Adaptation via Lower-Diag-Upper Decomposition for Parameter-Efficient Fine-Tuning
- 2026-09-28·PREreview·Cost-inefficient behaviors of coding agents: re-reads, subagents, test reruns, skill set size
Analyzing and Mitigating Cost-Inefficient Behaviors in Coding Agents
- 2026-09-26·PREreview·Routing coding agents across models: cache, context clearing, verification
Control the Harness, Control the Cost: Routing and Governing AI Coding Agents in the Enterprise
- 2026-09-25·Qeios·Rule-based LLM hallucination detector, re-review
- 2026-09-25·Qeios·URL metadata conditioning in LM pre-training
Metadata Conditioning Accelerates Language Model Pre-training
- 2026-09-25·Qeios·Label-free routing across multiple LLMs
- 2026-09-25·Qeios·Crosscoders for base-vs-chat model diffing
Overcoming Sparsity Artifacts in Crosscoders to Interpret Chat-Tuning
- 2026-09-25·Qeios·LLMs on symbolic time series (ABBA)
LLM-ABBA: Understanding Time Series via Symbolic Approximation
- 2026-09-25·PREreview·Memory injection cost in multi-agent workflows: caching, long runs, grading
Total Cost of Agency: Exact Attribution of Memory Injection Cost in Multi-Agent LLM Workflows
- 2026-09-23·Qeios·Hamiltonian mechanics for multi-hop reasoning chains
- 2026-09-23·Qeios·Retrieval-augmented QA over thousand-document haystacks
Document Haystacks: Vision-Language Reasoning Over Piles of 1000+ Documents
- 2026-09-23·Qeios·Multimodal LLM hallucinations on synthetic videos
VideoHallu: Evaluating and Mitigating Multi-modal Hallucinations for Synthetic Videos
- 2026-09-23·Qeios·Contactless HRV to tell humans from AI agents
- 2026-09-23·Qeios·Telemedicine follow-up after bariatric surgery during COVID-19
- 2026-09-23·Qeios·Rule-based LLM hallucination detector
SAS Multimetric Tribunal: Structural Classification of Hallucination in AI-Generated Text
- 2026-09-23·Qeios·Fine-tuning Phi-3 for multiple-choice QA
Fine-Tuning PHI-3 for Multiple-Choice Question Answering: Methodology, Results, and Challenges
- 2026-09-23·Qeios·Multi-step training for transformer maze navigation
- 2026-09-21·PREreview·Harness design for coding agents: context, planning and tool set
- 2026-09-21·Qeios·LLM persona simulation of lay health-seeking: methodological audit
- 2026-09-21·Qeios·Document poisoning in RAG: small-model robustness
In RAG We Trust? Measuring Robustness of Retrieval-Augmented Generation Under Document Poisoning
- 2026-09-21·Qeios·Review of human-AI collaboration in scholarly writing
An AI-Human Collaborative Review of AI-Human Collaborative Reviews
- 2026-09-21·Qeios·Tool for end-user auditing of generative AI
WeAudit: Scaffolding User Auditors and AI Practitioners in Auditing Generative AI
- 2026-09-21·Qeios·User-customizable ethical guardrails for LLM systems
AI Ethics by Design: Implementing Customizable Guardrails for Responsible AI Development
- 2026-09-21·Qeios·Position survey of AI for software engineering
- 2026-09-18·Qeios·Middleware for auditable, logged LLM interaction
MiND: A Minimal Deterministic Middleware for Auditable LLM Interaction
- 2026-09-18·Qeios·Indexing similar and related information for multihop retrieval
SiReRAG: Indexing Similar and Related Information for Multihop Reasoning
- 2026-09-18·Qeios·Survey of reinforcement learning for code generation
Enhancing Code LLMs with Reinforcement Learning in Code Generation: A Survey
- 2026-09-18·Qeios·Survey of vision-language models and their benchmarks
Benchmark Evaluations, Applications, and Challenges of Large Vision Language Models: A Survey
- 2026-09-18·Qeios·Cache mechanism for lifelong multi-agent path finding
Enhancing Lifelong Multi-Agent Path Finding with Cache Mechanism
- 2026-09-18·Qeios·Geometric taxonomy of hallucinations in language models
- 2026-09-18·PREreview·Agents reporting review work they did not perform
- 2026-09-17·Qeios·Mixed-precision KV cache quantization in llama.cpp
Beyond Perplexity: The Quality–Memory Frontier of Native Mixed-Precision KV Caching in llama.cpp
- 2026-09-17·Qeios·LLM-judge arbitration for free-form QA evaluation
DAFE: LLM-Based Evaluation Through Dynamic Arbitration for Free-Form Question-Answering
- 2026-09-17·Qeios·LLM confidence measures for social-science annotation
LLM Confidence Evaluation Measures in Zero-Shot CSS Classification
- 2026-09-17·Qeios·Medical chatbot combining retrieval and long-term memory
Hybrid Memory-Retrieval Model: Enhancing Trust in Medical Chatbots
- 2026-09-17·Qeios·Eliminating RAG hallucinations by rewriting queries and passages
- 2026-09-17·Qeios·Open-source LLMs generating synthetic toxicity data
ToxiLab: How Well Do Open-Source LLMs Generate Synthetic Toxicity Data?
- 2026-09-17·Qeios·Multi-agent LLM system for discovering physical laws
PhysAgent: A Multi-Agent Approach to the Automated Discovery of Physical Laws
- 2026-09-17·Qeios·Structured prompt design for LLM graph learning
GraphICL: Unlocking Graph Learning Potential in LLMs through Structured Prompt Design
- 2026-09-17·Qeios·Generation-augmented retrieval for fine-grained localization
- 2026-09-17·PREreview·Subagents versus skills: context isolation and its cost
Subagents vs Agent Skills: Executing Reusable Knowledge for Long-Horizon Agentic Tasks
- 2026-09-16·Qeios·Reproducibility of LLM personas in simulated health-seeking
Simulating Lay Health-Seeking Behavior with LLM Personas and Illness Vignettes
- 2026-09-16·Qeios·A multi-turn conversational RAG benchmark
MTRAG: A Multi-Turn Conversational Benchmark for Evaluating Retrieval-Augmented Generation Systems
- 2026-09-16·Qeios·Cost and quality of code agents on lightweight models
Matching Frontier Code Agents with Lightweight Models via Multi-Model Consultation
- 2026-09-16·Qeios·Epistemic herding in distributed multi-agent search
The Cost of Consensus: Malignant Epistemic Herding in Distributed Multi-Agent Search
- 2026-09-16·Qeios·The economics of reusing stored KV cache
Towards More Economical Context-Augmented LLM Generation by Reusing Stored KV Cache
- 2026-09-16·Qeios·Physical AI agents and vertical architectures
Physical AI Agents: Integrating Cognitive Intelligence with Real-World Action
- 2026-09-16·Qeios·Model diversity in multi-agent debate
Diversity of Thought Elicits Stronger Reasoning Capabilities in Multi-Agent Debate Frameworks
- 2026-09-16·Qeios·Geometry of context engagement in RAG
Semantic Grounding Index: Geometric Bounds on Context Engagement in RAG Systems
- 2026-09-16·Qeios·Robustness of RAG under document poisoning
In RAG We Trust? Measuring Robustness of Retrieval-Augmented Generation Under Document Poisoning
- 2026-09-16·PREreview·Context trimming and the reliability of agentic workflows
- 2026-09-15·Qeios·Benchmarking medical AI agents in an EHR environment
MedAgentBench: A Realistic Virtual EHR Environment to Benchmark Medical LLM Agents
- 2026-09-15·Qeios·Decentralised coordination of multi-agent systems
AgentNet: Decentralized Evolutionary Coordination for LLM-based Multi-Agent Systems
- 2026-09-15·Qeios·Agent-driven selection of training data
Star-Agents: Automatic Data Optimization with LLM Agents for Instruction Tuning
- 2026-09-15·Qeios·Agents in research idea generation
- 2026-09-15·Qeios·Autonomous agents in verification code generation
DRC-Coder: Automated DRC Checker Code Generation Using LLM Autonomous Agent
- 2026-09-15·Qeios·Model ensembles for bibliography generation
Enhancing Annotated Bibliography Generation with LLM Ensembles
- 2026-09-15·Qeios·Agentic neural graph databases
- 2026-09-15·Qeios·Vertical agentic systems
Agentic Systems: A Guide to Transforming Industries with Vertical AI Agents
- 2026-09-15·Qeios·Agentic web navigation and RAG in engineering design
Accelerating Manufacturing Scale-Up Using Agentic Web Navigation and Retrieval-Augmented AI
- 2026-09-14·PREreview·Decision-making under uncertainty
- 2026-09-13·PREreview·Evaluation of medical language models
The widening evaluation gap in medical large language model research 2023 to 2026
- 2026-09-12·PREreview·Reproducibility of agentic systems
- 2026-09-12·Qeios·Domain-specific financial language model
- 2026-09-12·Qeios·Identity drift in LLM agents
Prompt Volatility: An Empirical Study of Identity Drift in LLM Agents
- 2026-09-12·Qeios·Dynamic routing between models
- 2026-09-12·Qeios·Machine-assisted mathematical proofs
- 2026-09-12·Qeios·Cognitive bias in model outputs
- 2026-09-12·Qeios·Energy management in the home
- 2026-09-12·Qeios·Cultural alignment of models
- 2026-09-12·Qeios·Long-form opinion summarisation
Research and open data
A preprint, technical reports and the datasets behind them.
A Suite That Does Not Discriminate: The Case for Result-Set Fields in Agent Conformance Records
A set of conformance checks can consist entirely of sound, correctly reported, individually falsifiable checks and still support no conclusion, because no check changed state across the conditions compared. That is a property of the set, and no per-check field can carry it. The paper proposes a result-set field recording whether any check discriminated and which ones; it documents the adjacent case in a published corpus of 36 coding-agent runs, where a verdict of 4,086 tests and zero failures held with zero variance across six conditions whose modelled cost differed by 25.3% (a figure computed with the incorrect prices of report versions 1.0 and 1.1; 33.5% at the corrected prices of version 1.2, DOI 10.5281/zenodo.22759217); it reports as negative the tempting explanation that longer sessions simply did less work; and it identifies a case the proposed field cannot answer — where the checks are authored by the run itself, check identity does not survive across runs and the comparison-level question is undefined rather than false.
No Request With a Known Crawler User-Agent Reached llms.txt: Fifteen Days of Logs at One Origin Server
Fifteen days of logs at one origin server: 46,155 requests presenting known search and AI crawler user-agents, none for llms.txt, while robots.txt was requested 698 times in the two days after the file went live. Corrects the numbers in draft-arsentev-llm-context-discovery-00.
AI Content Engine Economics: The Cost of Short-Form Video Production by AI Agents and by People
One company's AI content engine produced 1,497 short-form videos across 67 channels in 101 days for $4,313.76 in subscriptions: $0.52 per 1,000 organic views and $2.88 per video. The same volume from a human studio would take 42–78 people and $179,000–1.21 million a month; buying the reach from influencers would cost 3.8–57 times more. Built by one person with no technical background; code written by Claude agents. Company under NDA, figures and calculations only.
Clear Every Third Task: A Measured U-Curve in the Context Economy of Coding Agents
Corrected 2026-09-15. How often should an agent's context be cleared? 36 runs, six session-length policies, six replicates each, and no failure among the tests the agent wrote (4086 in total). Cost is U-shaped in session length: clearing after every task costs 33.5% more than clearing every third (p = 0.0022), while never clearing costs 9.7% more, which is not significant (p = 0.046, Holm-adjusted 0.14). Three, four and six tasks per session are indistinguishable, so the low region is a plateau. The U decomposes into cache writes falling and context per call rising, with 1-hour cache writes priced 20:1 against cache reads.
The Context Economy of Agentic LLM Sessions: Where the Money Actually Goes
Corrected 2026-09-15. 722 agent sessions, 150,902 model calls, 34.6 billion tokens. Context handling accounts for 87.8% of modeled cost and generation for 12.2%; by estimate, 80% of the spend comes from 2.9% of sessions. Includes two log-deduplication traps that change the answer by a factor of two in either direction, and the de-identified dataset.
AI Tools Radar: GitHub and Hugging Face projects selected by arsentev.ai (monthly)
490 open-source AI projects selected and reviewed by the arsentev.ai radar since June 2026, with daily GitHub and Hugging Face API snapshots of how they evolve after selection. A new version is released every month.
Context U-curve: 36 coding-agent runs under six context-clearing policies
Run-level token counters, modeled cost, wall clock and test outcomes for every run behind the U-curve report. Counters only - no prompts, no model output, no paths.
Context economy: modeled cost of 722 coding-agent sessions
Per-session model calls and modeled cost at list prices for the 722 sessions behind the Context Economy report (v1.1). Counters only - no prompts, no code, no project names.
Standards
Internet-Drafts at the IETF.
Agent Run Metrics: A JSON Interchange Format for Resource Accounting of AI Agent Runs
draft-arsentev-agent-run-metrics
Discovery and Retrieval of Publisher-Curated Context Files for Large Language Models
draft-arsentev-llm-context-discovery
Open-source tools
contextburn — a context-accounting meter for coding agents, MIT licensed.
Memberships and participation
Standards and scholarly-communication bodies.
W3C — Cognitive AI Community Group
Participant
W3C — Agent Conformance and Benchmarking Community Group
Participant
Member
Member
Member since September 2026
Research — peer-reviewed papers and the author's identifiers. · Press · About