← 8bitconcepts
Research

All research papers

Field-level analysis of enterprise AI programs — what actually compounds, what stalls, and why. Practitioner research written for engineering teams at Series B-D companies.

Looking for a guided tour? Open the Research Atlas →

Subscribe via RSS — a dedicated feed for research papers, auto-refreshed weekly.

Need this work inside your business?
Same hands that wrote these. We embed and ship the AI systems your team actually uses.
Work with us →
Papers
The Agentic Commerce Gap
AI agents are beginning to shop on behalf of users. Most online businesses have no architecture for this. The gap between agent-ready and invisible is about to become a revenue gap.
agentic-commerceagentsstrategy
Claude Code Tips: 12 Ways to Get More Out of Every Session
Practical tips for developers already using Claude Code. Model switching, CLAUDE.md, MCP wiring, /compact timing, scoped permissions, and the patterns that actually improve output quality.
claude-codetipsdeveloper-tools
Claude Code Hooks: Automate Guardrails, Logging, and Workflow Enforcement
Hooks let you run scripts on Claude Code lifecycle events — block dangerous commands, log every file write, trigger notifications on task completion, auto-lint after edits. Power-user configuration guide.
claude-codehooksautomation
Claude Code CLAUDE.md: The Practical Guide to Project Memory
CLAUDE.md gives Claude Code persistent project context across sessions. What to put in it, what makes it worse, and the one problem it cannot solve.
claude-codeai-codingdeveloper-tools
How to Use Claude Code: A Practical Guide for 2026
Install, configure, and run Claude Code for real development work. Setup, CLAUDE.md, MCP wiring, and the patterns that actually make it useful in production.
claude-codegetting-starteddeveloper-tools
Claude Code MCP Servers: How to Add, Configure, and Use Them
Complete setup guide for MCP servers in Claude Code. Command syntax, scope options for solo vs. team use (.mcp.json), environment variable passing, and the tools worth wiring in.
claude-codemcpdeveloper-tools
Claude Code for Teams: Shared Context, API Keys, and Cost at Scale
Running Claude Code on a team of 10 costs $200–400/month and breaks in three predictable ways. Here's the shared CLAUDE.md pattern and API key setup that actually works.
claude-codeteamsenterprise
Claude Code Pricing: What It Actually Costs in 2026
Claude Code is billed on API usage, not a flat monthly fee. What you'll actually pay per session, how to control costs, and when it's worth it vs Cursor or Copilot.
claude-codepricingdeveloper-tools
Claude Code vs GitHub Copilot: Which One Is Right for You?
GitHub Copilot is an inline completion engine. Claude Code is an agentic task runner. They're not really competing — here's when each wins.
claude-codegithub-copilotdeveloper-tools
Claude Code vs Aider: Two Terminal Agents, Different Philosophy
Both are terminal-based AI coding agents billing at API rates. Aider is open-source and model-agnostic with git-first automation. Claude Code has MCP, CLAUDE.md memory, and 1M-token context. Here's how to choose.
claude-codeaiderdeveloper-tools
Claude Code vs OpenAI Codex CLI: Side-by-Side Comparison (2026)
Two terminal-native AI coding agents from Anthropic and OpenAI. Compared on model quality, pricing, CLAUDE.md vs AGENTS.md, MCP support, OS-level sandboxing, and git workflow depth.
claude-codecodexdeveloper-tools
Claude Code vs Gemini CLI: Side-by-Side Comparison (2026)
Claude Code vs Google's open-source Gemini CLI. Compared on free tier access, Google Search grounding, Plan Mode, GEMINI.md config, hooks, MCP support, and 1M-token context depth.
claude-codegemini-clideveloper-tools
Claude Code vs Windsurf: Which AI Coding Tool Is Right for You?
Claude Code is a terminal agent. Windsurf is a full IDE with Cascade and background Devin agents. Both are agentic — here's how to pick the right one for the task.
claude-codewindsurfdeveloper-tools
Claude Code vs Cursor: Which One Should You Actually Use?
Claude Code is a terminal-based agentic tool. Cursor is an editor. They're not really competing — but you need to know which one fits the task you're doing right now.
claude-codecursordeveloper-tools
Claude Code Context Limit: Why It Breaks Mid-Task and the Fix
Claude Code's context window fills up on long sessions and large repos. What actually works for continuing without losing your session.
claude-codeai-codingdeveloper-tools
The Self-Testing Layer
Agentic businesses do not fail because agents make mistakes. They fail because mistakes do not become structure. A researched operating model for artifact scoring, feedback loops, evaluator calibration, audit trails, and regression systems.
agentsevaluationself-improvement
Your AI Is Moving Back Onto the Machine
The future of AI inference is hierarchy: cloud for frontier work, devices for the everyday intelligence layer close to private context.
ai-strategyon-device-aiplatform-shift
The Compounding Gap
The lead fast-moving companies build over slow movers in 2026 isn't linear — it compounds. By the time you notice, the lead is structural.
strategyvelocitycompetitive-advantage
The Context Wall
AI agents fail 97.5% of real work. The fix isn't coding — it's the four pieces of context infrastructure most teams have no path to building alone.
agentsreliabilityinfrastructure
The Foundation Trap
Every AI architecture decision in 2026 is a bet on which infrastructure layer survives 2027. The five upstream decisions most operators make without naming.
architecturevendor-lock-instrategy
The Expansion Tax
Companies cutting headcount with AI are misreading the signal. When execution costs drop 10x, the market expands. The cutters are ceding that territory.
strategyroigrowth
The Domain Advantage
The 20 years of operating expertise you've built is exactly what AI cannot replicate. Two ingredients of a working AI workflow — operators already have the harder one.
operatorsexpertisestrategy
The PNW AI Desert
1 of 25 named AI hiring hubs is in the Pacific Northwest. Operators in Vancouver WA / Camas / Portland / Tigard cannot hire local AI engineers — and don't need to.
pnwlocal-marketsmb
The Integration Tax
Model API costs are 10–20% of what AI actually costs to ship. Where the other 80% goes.
integrationtcoenterprise
Beyond the Prompt
The teams shipping reliable production agentic systems are not prompting harder — they moved through a specific engineering maturity ladder.
llmengineeringsystems-design
The Six Percent
88% of organizations use AI. Only 6% see meaningful returns. What McKinsey found in 2,000 companies across 105 countries.
adoptioncase-studiesbest-practices
The Mandate Trap
Shopify's AI mandate worked. Duolingo's didn't. Companies copying the Shopify memo template are learning the wrong lesson.
adoptionleadershipstrategy
The Measurement Problem
A company ran an AI system for eight months before discovering four months of silent degradation. Most have no better detection mechanism.
roimetricsevaluation
The Org Chart Problem
AI transformation fails because of where it sits in the org chart. Every placement encodes a ceiling.
adoptionorganizational-designchange-management
Shift Handoff Intelligence
100% information retention with AI-generated shift briefings vs. 40–60% with verbal handoffs. Pattern-detection gap is where preventable failures originate.
agentscontextoperations
The Guardrails Gap
Engineering teams spent 2023 and 2024 obsessing over what AI would say. In 2026, the threat has shifted — agentic systems are now taking action.
agentssafetygovernance
The Hallucination Budget
Most engineering teams ship LLM features with less testing rigor than they apply to a login form. Production hallucinations land on customer trust and legal risk.
llmreliabilityevaluation
The Agentic Accountability Gap
Enterprise teams spent three years learning how to stop AI from saying the wrong thing. Then they handed those same systems write-access to production.
agentsgovernanceaccountability
Research topics

Papers aggregated by theme — built for long-tail search and agent discovery.

Agentic AI in Production Enterprise AI ROI AI Governance AI Organizational Design Reliability & Evaluation
Subscribe

Two papers a week on what's actually happening inside enterprise AI programs.

Prefer a reader? RSS feed.