AI Intelligence Archive

The decision to rewrite For years, rewriting an existing codebase from scratch was...

The work should be easy to verify and continue after the conversation ends. An agent changes the...

In my last post, I wrote about moving Forem’s performance and maintenance work into autonomous loops...

At Glean, we tested a model that does not write text. It only makes decisions. In our tests, it...

At the start of this year, Kent C. Dodds gave a coding agent a link to a GitHub issue he had been...

Most developers working with coding agents still interact with them directly. They open a terminal,...

Do you need to read every line of AI-generated code to trust it? Not necessarily. You need to check...

Whether you’re trying to build a software factory or already running one, you’ve probably heard some...

Set up your development tools in Docker Sandboxes, control network access, and reuse your environment with snapshots and Dockerfile-based templates.

When AI agents write to and read from your database through legitimate API calls, the attack lives in...

Every unattended loop I run carries a cost I never priced until last week's audit, and it comes...

Last post: 13 isolation checks passed while a Unix socket sat open. Those checks weren't nothing....

We've gotten really good at building AI Agents. With so many frameworks, and higher level...

Over a hundred runs of a controlled experiment on my own repository: what makes ADRs actually reach a coding agent, what parts of a record matter, and the one sentence that changed the result.

A few years ago (three years at most), one of the biggest challenges QA Engineers faced was: how do...

An agent that pays for things needs a limit: a maximum price per call, a payee you expect, sometimes...

Most multi-agent frameworks are benchmarked on 70B+ cloud models. Because frontier models have...

I once watched two agents agree on every message and still destroy a workflow. They were both...

How I built a mobile-responsive global bridge-strike, axle-load, and flood-scour safety platform using Gemma 4, Google ADK, OSRM routing, and 4,809 real FHWA NBI + OSM bridges.

What building Brainy and MoneyBrain taught me about finding early users beyond my LinkedIn network, and what I want to try next.

LiteLLM Rust’s 7x overhead and 9x memory claims, checked against its public benchmark files: what...

The AI race is no longer just about generating better answers. It's increasingly about building...

Foundation models pretrained on large natural-image collections are increasingly adapted for medical image segmentation, in the expectation that they will replace models trained for each task. Most comparisons, however, use data from the t…

DeepLearning (DL)-based Computer-Aided Diagnostic (CAD) systems have shown great promise in supporting clinicians, but their performance heavily depends on the availability of large, well-annotated datasets. In medical lesion detection, th…

We present TorchCurves, a PyTorch library that treats vector-valued parametric curves as learnable primitives. Its central operation maps batches of scalar inputs to points on curves represented in B-spline or Legendre polynomial bases, wi…

Reinforcement learning with verifiable rewards (RLVR) produces rich training experience, yet outcome rewards reduce each attempt to a single number and leave most of this information unused. Training-time scaffolds, such as strategies or h…

LLM agents are increasingly used for open-ended, data-driven scientific discovery, yet most of them evaluate hypotheses pointwise, testing an empirical claim $h$ only against its complement $\neg h$. When evidence rejects $h$, the agent le…

Predicting how a ligand binds to its target is central to small-molecule drug discovery. Modern cofolding models can generate diverse candidate poses, but reliably ranking these candidates remains challenging. We introduce Pyrite, a pairwi…

Connectome-constrained models are often used to argue that a measured wiring diagram outperforms shuffled or rewired wiring, and hence that structure shapes function. Such comparisons can confound wiring with how the models are tuned: a co…

Agent harnesses have emerged as a promising mechanism for continual learning, alongside test-time parametric learning. However, the current benchmarks have several limitations: they generally focus on a single applied domain, serve each in…

Foundation models trained on large unlabeled corpora develop emergent capabilities, tasks that go beyond the training objective. We extend this paradigm to multimodal biology with Bonbon, a foundation model for molecular interactions train…

Last week I asked a teammate, "Is the build green?" He replied with four paragraphs. There was...

Deploying LLM-based analytics agents in enterprise settings requires evaluation frameworks that can reliably detect failures across complex, multi-tool workflows. We present a three-phase comparative study of three evaluation frameworks, e…

Business process automation runs the same Standard Operating Procedure (SOP) again and again on new inputs. Large Language Model (LLM) agents can follow an SOP written in natural language, but they call an LLM to choose every step of every…

Time series analytics has been widely applied across various domains. While existing time series foundation models (TSFMs) have demonstrated promising results, they still face two major challenges: 1) Most TSFMs mainly rely on stacking Tra…

Research ideation is iterative: researchers refine their ideas step by step based on feedback. Yet existing evaluations obscure this search process, making it difficult to understand how LLMs explore the space of research ideas and what li…

Static scanners read npm packages without running them, so they miss the postinstall that only fires at install time. I gave my AI agent one tool that runs the install in a disposable AWS CodeBuild sandbox and catches the curl | bash from …

Neural networks are increasingly studied not merely as predictive systems, but as complex systems whose internal computations are themselves objects of scientific inquiry. Recent progress in mechanistic interpretability (MI) has made it po…

In 2017, eight researchers published a paper with the cockiest title in computer science: Attention...

In this paper, we present 1020 Group's submission to the WMT26 shared task on automated quality evaluation. Our system, Pterodactyl, was engineered specifically for subtasks 2 and 3, that is, segment-level quality score prediction and the …