Stack
Which services to use and what they cost: model access, rented GPUs, coding agents and the pipeline around them.
Automate SEC filings research with a coding agent, from EDGAR to a finished panel
One typed goal makes Claude Code pull SEC filings, build a firm-quarter panel, run the checks and commit the result. Auto mode and /goal keep it working until the checks pass.
Agent-first coding: use the frontier tools and pay for them
Claude Code or OpenAI Codex, desktop app or command line. Generic and budget setups hide what the new tools can do.
Using an AI agent on licensed data without risking your campus's access
Most data licenses forbid putting the content into AI systems. An agent can still write and repair the whole pipeline if the licensed rows never enter its context.
Negotiating an AI contract for beginners: retention, zero data retention, training and models
Four clauses decide whether an AI service is safe for research data. They are retention, training, access and model, and a sales answer on any of them counts for nothing until it is in the contract.
API, MCP, CLI or computer use: how to connect an AI agent to your research tools
Use the lowest-level, most structured interface the job allows. That means SQL or an API for data, the command line for software, and the screen only when nothing else reaches it.
Automated research pipelines run in four fields, and you can point one at finance
Machine learning, mathematics, biomedicine and finance each have a system that runs most of the research loop. Two of them you can install today.
Claude Code explained, and why it beats an AI plugin in VS Code
Claude Code is a program that takes a task, runs the commands and checks the result. In six months it passed GitHub Copilot among professional developers.
Give your agent a memory: the instruction file and your first skill
Two plain text files turn an agent from a stranger into a research assistant who knows the project. The same content works in Claude Code and OpenAI Codex.
OpenRouter or AWS Bedrock: where a researcher should buy model access
OpenRouter is the quick way to try many models with one key. Bedrock is the one to use when the data is licensed or the university already buys from AWS.
Token management for researchers, and why to buy usage before you hire help
Cached input costs a tenth of the normal price and batch jobs cost half. Spend on usage for the people you have before you spend on support staff.
Vast.ai or RunPod: renting a GPU for research without overpaying or leaking data
Vast.ai is a marketplace with the lowest floor prices and the most variance. RunPod has fixed prices and a data-center tier. Which one depends on what is on the disk.
Why data vendors are scared, and how to deal with them as a researcher
Their interface is worth less every month and they cannot see what your agent does. That gives you more room at renewal than you have had in twenty years.
If the platform forbids agents, build the dataset yourself from the source
Filings, financial statements and earnings calls were public before any vendor packaged them. An agent can go back to the source, and the tools to do it are free.
Developers are leaving Copilot and editor plugins for Claude Code and Codex
In six months Claude Code went from 18% to 39% of professional developers and Codex from 3% to 16%, while Copilot and Cursor fell. Price, tools and computer use explain it.
Keep paying for call transcripts, but only for the history, the identifiers and the rights
An agent now does the cleaning, splitting and scoring that used to justify a data budget. What it cannot produce is history, identifiers and the right to use the text.
The one public review of WRDS's Ask AI chatbot finds a lookup tool with a six-question limit
Librarians in Singapore tested the beta and found it useful for locating variables, unreliable on first try and capped at six questions a session.
A paper-writing pipeline for Claude Code went from 6,400 GitHub stars to 51,000 in five months
academic-research-skills runs 32 agents across ten stages with two integrity gates. Its builder says the checks caught 15 fabricated citations in one real paper.