Articles
Practical guides to building and running AI agents: research, retrieval, monitoring, tool access and the controls that keep their work reviewable.
These guides explain how agent work gets done in practice: how an agent finds and checks information, how it uses tools, how it watches sources that change, and where a person should stay in the loop. Each one ends with something you can try or verify on your own material.
Start with the concepts
Research you can check
- Research agents that hold up
- Reducing hallucinations with grounding
- Evaluating search on your own queries
Choosing tools
- How to choose a web search API
- Choosing search for AI agents
- Choosing a deep research API
- Choosing a web scraping API
Monitoring and alerts
Sales and company research
- Web enrichment for sales data
- Finding companies that match your ideal customer profile
- Automated due diligence research
The full list follows below. For short definitions of the terms these guides use, see Learn; for notes on approvals, evidence and audit, see the journal.
74 resources
- Articles
A competitive intelligence pipeline where every claim has a source
Run competitive intelligence with agents: questions first, sources mapped to them, a cadence per signal, evidence on every claim and review before anything is shared.
research - Articles
A deep competitor analysis an agent can run and you can verify
Plan a one-off competitor analysis an agent can execute: a decision to inform, approved sources, a comparison table with evidence per cell and honest open questions.
research - Articles
A LangChain search tool with citations, injection guards and tests
How to define a search tool for a LangChain agent, shape results for the model, keep citations verifiable, defend against injected pages and test the wiring.
research - Articles
A repeatable deal sourcing workflow: criteria, evidence, review
How to build a repeatable sourcing workflow for acquisition or investment targets: written criteria, evidence for every candidate, human review before outreach.
research - Articles
Agent-built market research reports that survive review
Framing the question, approving sources, structuring the report and checking evidence before an agent-built market report or intelligence feed is shared.
artifact - Articles
AI agents, defined: goal, tools, loop and memory
A working definition of an AI agent built on goal, tools, loop and memory, how it differs from chatbots and workflows, and what makes it safe to delegate to.
research - Articles
Build or buy a research agent: a decision guide for team leads
How to choose between building a web research agent and buying a deep research API: control, evidence, upkeep, cost model and governance, plus a checklist.
research - Articles
Building a market map an agent can defend and keep current
Define the category, turn it into criteria an agent can test, collect candidates with evidence, report coverage honestly and keep the map current as markets move.
research - Articles
Chatbots that stay current: retrieval, freshness and when to refuse
How to design a chatbot or assistant that answers from current sources: the retrieval step, freshness rules, real-time data tools, citations and when to refuse.
chat - Articles
Choosing a deep research API: depth, speed, cost and citations
How to choose a deep research API: depth, speed and cost trade-offs, report structure, citation quality, faster modes, and an evaluation you run on your own questions.
research - Articles
Choosing a personal AI agent you can trust with real access
What to compare across self-hosted and hosted personal agents, how they tend to fail, and the scoping, approval and logging habits that keep them dependable.
cli - Articles
Choosing a search tool your agent can actually work with
What an agent needs from search: machine-readable results, passages, citations, freshness and tool-calling fit, plus a checklist for comparing options on your tasks.
research - Articles
Command-line interfaces, and why agents are so good at them
What a command-line interface is, why text-in, text-out tools suit AI agents so well, and the guardrails to set before an agent runs commands for you.
cli - Articles
Company and industry news alerts that stay worth opening
How to write news alert rules for single companies and whole industries, collapse duplicate stories, define materiality and choose between digests and real-time.
monitor - Articles
Company research data: registries, filings, news and vendors compared
A selection guide to company research data: registries, websites, news, filings and vendors, weighed by coverage, freshness and provenance, with a checklist.
research - Articles
Data enrichment: adding attributes you can trace back to a source
What data enrichment is, the common types and sources, why matching and provenance decide its quality, and how to keep enriched records correct over time.
research - Articles
Deep research by AI agents: how it works and how to review it
How agent-led deep research plans, gathers sources, synthesizes and cites; when it is worth using, how to brief it, and how to review the report it returns.
research - Articles
Detecting what a competitor really shipped, with evidence per change
How to tell announced features from shipped ones by watching changelogs, docs, API references and pricing pages against a baseline, with evidence for each change.
monitor - Articles
Due diligence and vendor checks that end in a human sign-off
Run due diligence and vendor checks with agents: answerable checklist items, evidence for each one, red flags defined in advance and a person who signs.
research - Articles
Enriching CRM accounts from the public web, field by field
Which CRM fields are worth enriching from public web sources, how to keep provenance per field, set a refresh cadence and review changes before the CRM sees them.
research - Articles
Enrichment APIs and a company database that stays correct
How to choose an enrichment API, store company data as sourced observations, resolve entities and conflicts, and refresh values so the database stays correct.
research - Articles
Enterprise deep research: requirements to settle before you buy
A selection guide for deep research inside a company: data boundaries, source control, citations, approval, audit and cost bounds, plus questions for any vendor.
research - Articles
Finding, enriching and qualifying leads without losing the evidence
An end-to-end lead pipeline from target definition to approved outreach: what each stage hands over, quality gates, consent and suppression, and where people decide.
research - Articles
Free and self-hosted web search for agents: what it really costs
A guide to free tiers and self-hosted metasearch for agents: the hidden costs, terms of use, reliability, privacy and the signals that it is time to pay.
research - Articles
Fresh web data in RAG: acquisition, freshness and citations
How to feed current web pages into a RAG pipeline: acquiring them, tracking freshness and provenance, chunking for citation, and refreshing without silent drift.
knowledge - Articles
From buying signal to first email: choosing and vetting trigger events
Which trigger events actually signal a need, how to cut alert noise, how to tie each event to the right account, and why outreach waits for a person's approval.
monitor - Articles
From ICP slide to checkable criteria and an evidenced match list
Turn an ideal customer profile into criteria a machine can check, collect matching companies with evidence for every field and report what the search missed.
research - Articles
Giving a local-model agent web access without leaking your queries
How to add web search to an agent running on Ollama or LM Studio: the realistic options, what each reveals, wiring the tool call, and safety rules.
research - Articles
Grounded answers: passage citations, refusals and support tests
How to ground model answers in retrieved sources: passage-level citations, refusing when nothing supports a claim, and tests that check every claim is backed.
research - Articles
Hosted search tool or your own retriever in an agent framework
Built-in web search or your own backend in an agent framework, with OpenAI's Responses API as the example: control, citations, cost model and how to evaluate.
chat - Articles
How agent skills work: structure, loading, portability and limits
What an agent skill is, how a SKILL.md folder is laid out, when the agent loads it, how portable it is, and why a skill never grants permission on its own.
cli - Articles
How an agent system fits together and where control belongs
The parts of an agent system (loop, tools, memory, evidence, approval, audit), common patterns, where web access plugs in and why control sits outside the model.
research - Articles
How web crawlers find, fetch and respect the pages they visit
How crawlers discover URLs, fetch pages without overloading sites, follow robots rules, identify themselves, and what happens to the data they collect.
knowledge - Articles
Inside a search index: building it, refreshing it, knowing its limits
How crawled pages become a searchable index, how inverted and vector indexes differ, and why coverage, freshness and cost always pull against each other.
knowledge - Articles
Inside the harness: the runtime that turns a model into an agent
What an agent harness does around a language model (loop, tools, context, permissions, limits) and why it shapes behavior as much as the model itself.
cli - Articles
Knowledge-work agent or coding agent: where Cowork and Claude Code fit
For operations teams: when a desktop knowledge-work agent like Claude Cowork fits, when a coding agent like Claude Code does, and how to scope access and review.
cli - Articles
Making a chatbot show its evidence, one passage at a time
How to build citation into a chatbot: ground answers in retrieved passages, display sources well, handle unsupported claims and test that each citation holds up.
chat - Articles
Meaning-based retrieval: embeddings, hybrid ranking and evaluation
How embeddings let search match meaning rather than exact words, where keyword search still wins, how hybrid search merges both, and how to evaluate results.
knowledge - Articles
Migrating off a search API: inventory, field mapping and a safe cutover
A plan for leaving a search API you depend on: inventory every usage, map result fields, evaluate side by side on real traffic and cut over with a way back.
research - Articles
Model Context Protocol explained: roles, primitives and risks
How the Model Context Protocol links AI applications to tools and data: hosts, clients and servers, tools, resources, prompts, transports and the security questions.
cli - Articles
Monitoring portfolio and watchlist companies against your thesis
How to turn an investment thesis into signals worth monitoring, grade sources before trusting them, and keep a person between a detected change and a conclusion.
monitor - Articles
Monitoring web pages for change: baselines, noise and review
How to choose and set up web change monitoring, including near-real-time feeds for agents: baselines, materiality rules, noise control and a review step.
monitor - Articles
One agent, three jobs: separating research, monitoring and extraction
How to design an agent that researches, monitors and extracts without mixing them up: separate plans and outputs, one evidence store, approvals at the boundary.
monitor - Articles
OpenClaw or Claude Code: two agents built for different jobs
How a self-hosted personal assistant like OpenClaw and a coding agent like Claude Code differ in the jobs they suit, the access they need, their risks and upkeep.
cli - Articles
OpenCode or Claude Code: weighing model choice, extensions and governance
Compare an open-source coding agent (OpenCode) with a vendor one (Claude Code) on model choice, extensibility, support and governance, then run your own pilot.
cli - Articles
Outgrowing keyword alerts: what a real replacement needs
Why keyword alerts miss what matters and bury it in duplicates, and what to require from a replacement: source control, meaning-based rules, dedup, evidence, APIs.
monitor - Articles
Picking a web search API for AI: a scoring sheet you run yourself
How to compare web search APIs for AI applications: result types, own index vs metasearch vs SERP scraping, freshness, terms, cost model and a scoring sheet.
research - Articles
Pulling clean article text and metadata out of web pages
How to strip boilerplate, recover authors and dates you can trust, respect robots rules and licensing, and keep a source reference on every extracted article.
knowledge - Articles
Reading technology use from public signals, with confidence levels
Which public signals reveal the technology a company runs, how to grade confidence, where false positives come from and how to automate detection with review.
research - Articles
Recall and accuracy in agent search: how to measure them yourself
What recall and accuracy mean when an agent does the searching, why vendor figures rarely line up, and a method to measure both on your own task.
research