# The Unblocked Library

URL: https://getunblocked.com/library/

In-depth guides, comparisons, and resources on AI coding agents, developer productivity, and engineering tooling.

---

- [AI-Powered Legacy Code Modernization: The Complete Guide (2026)](https://getunblocked.com/blog/legacy-code-modernization/) — Legacy code modernization with AI fails on missing context, not weak models. Meta took agent-usable coverage from 5% to 100%. The 2026 playbook for leaders. _(2026-06-30T08:00:00Z)_
- [Claude Code's Context Window: How It Works and How to Manage It (2026)](https://getunblocked.com/blog/claude-code-context-window/) — Claude Code's context window is 200,000 tokens by default, 1M on Opus 4.6 and later. Size isn't the lever. See where it goes and how to manage it. _(2026-06-25T08:00:00Z)_
- [Jira MCP: How to Connect Jira to Your AI Coding Agent (2026)](https://getunblocked.com/blog/jira-mcp/) — Connect Jira to your AI agent in ~2 minutes via Atlassian's official remote MCP server, then handle the 2 things every walkthrough skips. _(2026-06-24T08:00:00Z)_
- [AI Coding Anti-Patterns: How Teams Waste the Three Levers](https://getunblocked.com/blog/ai-coding-anti-patterns/) — Devs were 19% slower with AI yet felt 20% faster. The 7 AI coding anti-patterns that corrupt cost, output, and autonomy, plus the fix for each. _(2026-06-23T08:00:00Z)_
- [How Much Autonomy Should You Give Your AI Coding Agents?](https://getunblocked.com/blog/ai-coding-agent-autonomy/) — Autonomy isn't a slider you crank to 'more.' Set it to the weaker of two axes. Gartner predicts over 40% of agentic AI projects get canceled by 2027. _(2026-06-22T08:00:00Z)_
- [Why AI Agents Burn Tokens: The Four Runtime Waste Mechanisms](https://getunblocked.com/blog/why-ai-agents-burn-tokens/) — AI agents burn tokens four ways once they start working. Input tokens average 53.9% of agentic spend, mostly the agent talking to itself. Here is the signal that catches each waste mechanism, and the fix. _(2026-06-15T10:00:00Z)_
- [FinOps for AI Coding: Who Owns the Budget When Agents Spend It](https://getunblocked.com/blog/finops-for-ai-coding/) — Uber burned its entire 2026 AI coding budget in about four months and capped engineers at $1,500/month. FinOps for AI coding is the discipline that prevents that surprise. _(2026-06-15T09:30:00Z)_
- [Building With AI Effectively: An Engineering Leader's Operating Model](https://getunblocked.com/blog/building-with-ai-effectively/) — Only ~39% of organizations report measurable EBIT impact from AI. Building with AI effectively isn't a tool choice. It's an operating model built on three levers. _(2026-06-15T09:00:00Z)_
- [Tokenmaxxing Is Over. Token Yield Is the Scoreboard, and Context Moves It.](https://getunblocked.com/blog/token-yield-context-problem/) — GitHub Copilot's June 1 token billing sent agentic bills up 10-50x. The fix isn't cheaper tokens. It's token yield, the useful output you get per token, and context is the lever that moves it. _(2026-06-15T08:00:00Z)_
- [Cost per Merged PR: The Unit Economics of AI Coding](https://getunblocked.com/blog/cost-per-merged-pr/) — AI-assisted PRs run about 18.2% larger and review time can climb 91%, so cost-per-token looks cheap while the cost per merged PR quietly rises. _(2026-06-12T08:00:00Z)_
- [How to Cut AI Token Costs Without Throttling Your Agents: 7 Techniques](https://getunblocked.com/blog/reduce-ai-token-costs/) — You can reduce AI token costs 50-90% without throttling your agents by cutting the tokens that never contributed to output, not by capping capability. _(2026-06-11T08:00:00Z)_
- [The Auto-Loop Tax: Why Self-Running AI Agents Burn More Tokens, Not Fewer](https://getunblocked.com/blog/agent-auto-loop-token-cost/) — AI lab leaders are selling self-running loops as efficiency. Anthropic's own research shows multi-agent systems burn roughly 15x the tokens of a chat, and longer loops get worse, not cheaper. _(2026-06-10T08:00:00Z)_
- [Model Routing for Coding Agents: When Cheaper Models Save Money (and When They Cost More)](https://getunblocked.com/blog/model-routing-coding-agents/) — Model routing can cut a coding-agent bill 30-40%, then watch rework erase the win. Here is where the savings are real and where a cheap mis-route costs more. _(2026-06-09T08:00:00Z)_
- [AI Tokenomics: How to Measure and Control What AI Coding Tools Actually Cost](https://getunblocked.com/blog/ai-tokenomics-cost-framework/) — Enterprise LLM spend doubled from $3.5B to $8.4B in six months, yet only 39% of organizations report any EBIT impact. The bills are about yield, not price. _(2026-06-08T08:00:00Z)_
- [How to Measure the ROI of AI Coding Tools (When Fewer Than 1 in 5 Teams Do)](https://getunblocked.com/blog/measure-ai-coding-tool-roi/) — Atlassian found 99% of developers save time with AI, yet 50% still lose 10+ hours a week. Here is the formula that nets it out. _(2026-06-05T08:00:00Z)_
- [DORA Metrics in the AI Era: Necessary but No Longer Sufficient](https://getunblocked.com/blog/dora-metrics-ai-era/) — 90% of developers now use AI at work, yet the 2025 DORA report shows delivery stability still drops. Here's how to read the four keys. _(2026-06-04T08:00:00Z)_
- [Context-Adjusted Productivity: Measuring What AI Actually Ships in Large Codebases](https://getunblocked.com/blog/context-adjusted-productivity/) — AI lifts merged PRs 98% but adds 91% to review time. Context-adjusted productivity is the number that survives that math. _(2026-06-03T08:00:00Z)_
- [The AI Productivity Paradox: Why Your Devs Feel Faster but Ship the Same](https://getunblocked.com/blog/ai-productivity-paradox/) — Experienced developers were 19% slower with AI yet believed they were 20% faster. That 39-point gap is the AI productivity paradox. _(2026-06-02T08:00:00Z)_
- [How to Measure AI Productivity: The 2026 Framework for Engineering Leaders](https://getunblocked.com/blog/how-to-measure-ai-productivity/) — Experienced developers were 19% slower with AI yet believed they were 20% faster. Here is how to measure AI productivity across four dimensions. _(2026-06-01T08:00:00Z)_
- [How to Scale from One Agent to Parallel Agents Without Losing Quality](https://getunblocked.com/blog/scale-parallel-ai-agents/) — Multi-agent coordination improves parallelizable tasks by 81% but degrades sequential ones by up to 70%. Here are the coordination patterns that preserve quality. _(2026-05-29T08:00:00Z)_
- [Memory MCP Servers Compared: Stash, MemPalace, Hindsight, agentmemory](https://getunblocked.com/blog/memory-mcp-servers-compared/) — 4 memory MCP servers benchmarked on LongMemEval, BEAM, token cost, and architecture. MemPalace loads in 170 tokens. Hindsight leads at 10M-token scale. _(2026-05-27T08:00:00Z)_
- [How to Build a Context Layer for Your Engineering Org](https://getunblocked.com/blog/build-context-layer-engineering/) — A 5-step implementation guide for engineering leads building a context layer. Audit sources, choose architecture, connect systems, validate quality, iterate. _(2026-05-26T10:00:00Z)_
- [The Context Infrastructure Stack: A CTO's Guide](https://getunblocked.com/blog/context-infrastructure-stack-guide/) — Context infrastructure is the platform layer between data systems and AI agents. This CTO's guide maps the 4-layer stack powering production AI. _(2026-05-26T08:00:00Z)_
- [MCP Tool Overload: A Measured Guide to Context-Window Bloat](https://getunblocked.com/blog/mcp-tool-overload/) _(2026-05-16T08:00:00Z)_
- [Unblocked vs Glean vs Augment: Which Is Right for Your Engineering Team?](https://getunblocked.com/blog/unblocked-vs-glean-vs-augment/) — Glean, Augment, and Unblocked are often grouped together as “AI for engineering,” but they solve different problems. Glean is enterprise search, Augment is an IDE-native coding assistant, and Unblocked is engineering context infrastructure for humans and AI agents. _(2026-05-16T08:00:00Z)_
- [The 42,000-Token Problem With the GitHub MCP Server (and 4 Fixes)](https://getunblocked.com/blog/github-mcp-token-cost/) — The official GitHub MCP server charges ~42,000 tokens in tool definitions before any work begins. Four measured fixes, including Claude Code 2.0's tool-search subagent, cut the bill 25-50%. _(2026-05-15T15:00:00.000Z)_
- [The Context Layer: What It Is and Why Every AI Agent Needs One](https://getunblocked.com/blog/context-layer-ai-agents/) — A context layer sits between an AI agent and every system where its institutional knowledge lives. Here's what a real one includes, why MCP alone isn't one, and the four properties that distinguish it from a tool registry. _(2026-05-15T13:00:00.000Z)_
- [Why Your AI Coding Agent Keeps Forgetting Your Codebase](https://getunblocked.com/blog/claude-code-forgets-codebase/) — Engineers spend the first 5 minutes of every Claude Code session re-explaining their stack. Memory MCP servers help with fact recall but miss the bigger 'shape forgetting' pattern. Here's the structural fix. _(2026-05-14T15:00:00.000Z)_
- [Where Your MCP Token Budget Actually Goes: A 2026 Autopsy](https://getunblocked.com/blog/mcp-token-budget-autopsy/) — The GitHub MCP charges ~42,000 tokens before any real work begins. This is a line-item autopsy of the context tax, priced against the CLI alternative, plus four patterns that cut the bill. _(2026-05-14T13:00:00.000Z)_
- [Why Claude Code Keeps Reinventing Your Team's Rejected Approaches](https://getunblocked.com/blog/claude-code-institutional-memory/) — The trial-and-error tax in Claude Code is a memory problem, not a model problem. Here's the taxonomy of agent memory and the retrieval pattern that fixes it. _(2026-05-02)_
- [Claude Code Context Rot: Why Accuracy Drops at 200K Tokens](https://getunblocked.com/blog/context-rot-claude-code/) — Context rot is the measurable accuracy drop that hits every frontier LLM as input length grows. Here's how it shows up in Claude Code, and five retrieval patterns that recover most of what you lose past 200K tokens. _(2026-05-02)_
- [Three Hard Lessons from Building Context at Scale](https://getunblocked.com/blog/three-hard-lessons-context-scale/) — Three things we got wrong before we got our context engine right: optimizing for access instead of understanding, hiding conflicts instead of surfacing them, and caching answers instead of computing them. The retrospective behind the talks I gave at QCon London and AI Engineering London. _(2026-05-01T08:00:00Z)_
- [The AI Coding Tool ROI Reckoning: When Finance Starts Reconciling Spend with Output](https://getunblocked.com/blog/ai-coding-tool-roi-reckoning/) — AI coding tool spend is heading for a CFO conversation that hasn't happened yet. Worldwide AI spending will hit $2.5T in 2026, but 95% of GenAI pilots show no measurable P&L impact. Here is the spend trajectory, the output gap, and what engineering leaders should do before the audit. _(2026-04-30T08:00:00Z)_
- [Code Plus MCP Isn't a Context Engine: What a Real Reasoning Layer Actually Ingests](https://getunblocked.com/blog/code-isnt-context-mcp-isnt-enough/) — An AI coding agent wired to your repo and a few MCPs is still missing most of the context it needs. The decisions, conventions, and history that explain why the code is the way it is don't live in the code. Here's what a real context engine ingests, and why. _(2026-04-30T08:00:00Z)_
- [Building a Context Engine on Top of MCP: What's Possible and What's Missing](https://getunblocked.com/blog/building-a-context-engine-on-mcp/) — MCP is a transport protocol, not a context engine. Here's how to layer reasoning, conflict resolution, and permissions above your MCP servers — three patterns that work, four gaps the spec doesn't fill, and a build-vs-buy rubric. _(2026-04-29T08:00:00Z)_
- [The 8 Levels of Agentic Engineering: An AI Maturity Model for Engineering Teams](https://getunblocked.com/blog/8-levels-of-agentic-engineering/) — Engineering teams are scattered across an 8-stage AI maturity curve, from tab complete to autonomous agent teams. The infrastructure each stage needs is fundamentally different. Find your level and the right next step. _(2026-04-28T08:00:00.000Z)_
- [Context Engine vs Search Layer vs Orchestration Layer: The Real Split](https://getunblocked.com/blog/context-engine-vs-search-vs-orchestration/) — Three categories of AI infrastructure are being sold under the name 'context engine' in 2026, and the conflation produces predictable failure. Here is what each layer actually does, why they are complementary, not substitutes, and how to know which one you need. _(2026-04-28T08:00:00.000Z)_
- [The 5 Criteria That Define a Context Engine (And Which Tools Meet Them)](https://getunblocked.com/blog/context-engine-criteria-rubric/) — Everyone markets a 'context engine' now. Very few tools actually meet the bar. Here's the 5-criteria rubric that separates real context engines from retrieval layers, plus a scorecard of how every major tool performs. _(2026-04-27T08:00:00.000Z)_
- [Same Prompt, Same Model, Different Context: A Controlled Test](https://getunblocked.com/blog/same-prompt-same-model-different-context/) — In a controlled A/B test, an AI coding agent with Unblocked context cut tokens by 42%, ran 27% faster, made 64% fewer tool calls, and scored 41/50 vs 24/50 from a blind LLM judge. Here's the methodology and how to reproduce it. _(2026-04-27T08:00:00.000Z)_
- [How to Implement Context Engineering for AI Coding Agents](https://getunblocked.com/blog/how-to-implement-context-engineering/) — A step-by-step context engineering implementation guide: audit agent failures, map knowledge sources, build retrieval, enforce permissions, and measure results. _(2026-04-24T10:00:00Z)_
- [Context Engine vs Knowledge Graph: When You Need Understanding, Not Just Data](https://getunblocked.com/blog/context-engine-vs-knowledge-graph/) — Knowledge graphs map relationships between entities. Context engines reason across sources to produce answers. Here is when each one fits and where graphs alone fall short. _(2026-04-24T08:00:00Z)_
- [How a Context Engine Actually Works: An Architecture Deep Dive](https://getunblocked.com/blog/how-a-context-engine-actually-works-and-why-you-need-to-care-now/) — A working context engine is a five-stage system: ingestion, graph construction, retrieval-reasoning loop, conflict resolution, and permissioned delivery. Here is what each stage actually does and where the hard engineering sits. _(2026-04-23T12:00:00Z)_
- [How to Evaluate a Context Engine: The Buyer's Checklist](https://getunblocked.com/blog/how-to-evaluate-a-context-engine/) — A context engine evaluation should answer one question: will this system give your agents and engineers decision-grade context, or just better search results? This checklist covers data sources, conflict resolution, permissions, MCP delivery, and the retrieval-to-reasoning pipeline. _(2026-04-23T10:00:00Z)_
- [Unblocked vs Glean: The Context Engine for Engineering Teams](https://getunblocked.com/blog/unblocked-vs-glean/) — Unblocked is the context engine purpose-built for engineering teams. See why teams at Clio, TravelPerk, HeyJobs, Drata, RB Global, and Subsplash chose it over Glean for AI-agent and engineering context. _(2026-04-23T08:00:00Z)_
- [When to Use MCP vs CLI for AI Agent Tools (2026 Decision Guide)](https://getunblocked.com/blog/when-to-use-mcp-vs-cli/) — MCP and CLI aren't competitors, they're different surfaces for different agents. A practical 6-factor rubric for picking the right one (or shipping both). _(2026-04-22T20:00:00Z)_
- [Unblocked vs Sourcegraph Cody: Code Context for Enterprises](https://getunblocked.com/blog/unblocked-vs-sourcegraph-cody/) — Developers spend 60% of their time reading code, not writing it. Compare Sourcegraph Cody and Unblocked on data sources, context depth, and enterprise fit. _(2026-04-22T10:00:00Z)_
- [Three Myths About Context for AI Agents (And What Actually Works)](https://getunblocked.com/blog/three-myths-about-context-for-ai-agents/) — Naive RAG, stacking MCP servers, and bigger context windows are the three fixes teams reach for when AI agents ship bad code. All three fall short. Here's what the research says, and what actually works. _(2026-04-22T08:00:00Z)_
- [Context Engineering: The Complete Guide for Engineering Leaders](https://getunblocked.com/blog/context-engineering-guide/) — Why AI coding tools keep shipping rejected PRs, and how context engineering, the discipline behind the WHY of code, fixes it. A pillar guide for engineering leaders. _(2026-04-21T12:00:00Z)_
- [Context Engineering vs RAG: When to Use Which Approach](https://getunblocked.com/blog/context-engineering-vs-rag/) — RAG retrieves similar documents. Context engineering decides what an agent should know before it acts. Here is when each approach fits, when they overlap, and how to choose. _(2026-04-21T10:00:00Z)_
- [ServiceNow vs Tabnine vs Unblocked: Enterprise Context Engines Compared](https://getunblocked.com/blog/servicenow-vs-tabnine-vs-unblocked/) — Three vendors call their product a context engine, but they mean different things. This enterprise context engine comparison breaks down ServiceNow, Tabnine, and Unblocked across architecture, sources, permissions, and pricing so you can pick by fit. _(2026-04-21T08:00:00Z)_
- [What Is a Context Engine? A Complete Guide for Engineering Teams](https://getunblocked.com/blog/what-is-a-context-engine/) — A direct-answer guide to what a context engine is, how it differs from RAG and a context layer, and when engineering teams actually need one. _(2026-04-20T12:00:00Z)_
- [Context Engine vs Enterprise Search: Glean, Elasticsearch, and the Limits of Search](https://getunblocked.com/blog/context-engine-vs-enterprise-search/) — Enterprise search finds documents. A context engine resolves conflicts, enforces permissions, and reasons across sources. Here is where each one fits and where search stops being enough. _(2026-04-18T12:00:00Z)_
- [The AI Agent Doom Loop: Why Your Agent Keeps Making the Same Mistake](https://getunblocked.com/blog/ai-agent-doom-loop/) — The AI agent doom loop isn't a model problem. It's a context problem. Here's why coding agents repeat the same mistakes, and what actually breaks the cycle. _(2026-04-18T10:00:00Z)_
- [Satisfaction of Search: Why AI Agents Stop Before Finding the Right Answer](https://getunblocked.com/blog/satisfaction-of-search/) — Satisfaction of search is a fifty-year-old cognitive bias from radiology. It explains why AI coding agents stop at the first plausible result and ship code that compiles but fails review. _(2026-04-18T08:00:00Z)_
- [You Are the Context Engine (And That Doesn't Scale)](https://getunblocked.com/blog/you-are-the-context-engine/) — Engineers have become the manual context middleware between AI agents and their organizations. Here's why that collapses, what it costs, and what has to change. _(2026-04-17T12:00:00Z)_
- [Context Engineering vs Prompt Engineering: What's the Difference?](https://getunblocked.com/blog/context-engineering-vs-prompt-engineering/) — Prompt engineering shapes how you ask. Context engineering shapes what the model knows. Here's where each discipline starts, stops, and how they work together in production. _(2026-04-17T10:00:00Z)_
- [Decision-Grade Context: Why Retrieval Isn't Enough for AI Agents](https://getunblocked.com/blog/decision-grade-context/) — Decision-grade context is context an AI agent can act on without re-verification: synthesized across sources, conflict-resolved, and permission-enforced. Here's why retrieval alone can't get there. _(2026-04-17T08:00:00Z)_
- [Why MCP Servers Alone Aren't Enough for Enterprise Context](https://getunblocked.com/blog/why-mcp-servers-arent-enough/) — MCP is a solid transport layer for AI agents, but plugging in 12 servers doesn't solve enterprise context. Here's what's missing above the wire, and why reasoning beats retrieval every time. _(2026-04-16T11:00:00Z)_
- [Stop Babysitting Your Agents: How to Get Mergeable Code on the First Run](https://getunblocked.com/blog/stop-babysitting-your-agents/) — Most engineers are still the context engine for their AI agents. Here's why that collapses at parallel scale, and what mergeable code on the first run actually requires. _(2026-04-16T10:00:00Z)_
- [What Is Context Engineering? Definition, Principles, and Why It Matters](https://getunblocked.com/blog/what-is-context-engineering/) — Context engineering is the discipline of curating, retrieving, and governing the information an AI system needs to answer correctly. Here is the working definition, 5 principles, and why it matters now. _(2026-04-16T09:00:00Z)_
- [Context Engine vs RAG: Why Retrieval Alone Isn't Enough](https://getunblocked.com/blog/context-engine-vs-rag/) — RAG retrieves documents. A context engine resolves conflicts, enforces permissions, and reasons about what retrieval returned. Here is when each one fits. _(2026-04-16T08:00:00Z)_
- [The Context Problem Moved From People to Agents](https://getunblocked.com/blog/the-context-problem-moved-from-people-to-agents/) — Engineering teams have always had a context problem. New hires took months to ramp. Tribal knowledge lived in a few heads. Now that same problem lives in every AI agent session, and the old workarounds don't work. _(2026-03-30T17:29:50.000Z)_