NexcurAI - Writing Essays on operating with Claude inside a consulting firm. https://nexcur.ai/blog/ 2026-04-22T00:00:00Z Copyright 2026 NexcurAI Inc. NexcurAI Editorial hello@nexcur.ai https://nexcur.ai/favicon.svg https://nexcur.ai/og-image.png The self-critique loop: how we get Claude to find its own mistakes tag:nexcur.ai,2026-04-22:/blog/self-critique-loop 2026-04-22T00:00:00Z 2026-04-22T00:00:00Z the operator Every handbook runs through a self-critique pass before a human sees it. Prompt patterns and real traces from our production pipeline. Threat modeling goes from 3 days to 3 hours when Claude drives STRIDE tag:nexcur.ai,2026-04-21:/blog/threat-modeling-3-hours 2026-04-21T00:00:00Z 2026-04-21T00:00:00Z the operator STRIDE with Claude driving the enumeration and a human driving the judgment. A workshop trace that fits into one afternoon. What tool use actually unlocks tag:nexcur.ai,2026-04-19:/blog/tool-use-unlocks 2026-04-19T00:00:00Z 2026-04-19T00:00:00Z the operator Tool use is the feature that turns Claude from an essay writer into a collaborator that can touch real systems. What it unlocks, what it does not, and the design discipline that keeps tool-using agents from becoming a liability. The 200k window is a context engineering problem, not a memory problem tag:nexcur.ai,2026-04-19:/blog/200k-context-engineering 2026-04-19T00:00:00Z 2026-04-19T00:00:00Z the operator The 200k context is a staging area you engineer, not memory. Retrieval patterns for long-running engagements. How to build a topical map that ranks in Google AND gets cited by ChatGPT tag:nexcur.ai,2026-04-18:/blog/topical-map-dual-optimization 2026-04-18T00:00:00Z 2026-04-18T00:00:00Z the operator The three-layer topical map (hub, spokes, citation surfaces) that serves both Google SERPs and answer-engine citations. Why fixed-scope pricing is 10x better tag:nexcur.ai,2026-04-17:/blog/pricing-fixed-scope-10x 2026-04-17T00:00:00Z 2026-04-17T00:00:00Z the operator Fixed scope as an operating discipline. The math and the four change-request clauses that make it survive contact with reality. The incident response runbook nobody wrote because nobody had time tag:nexcur.ai,2026-04-16:/blog/incident-response-runbook 2026-04-16T00:00:00Z 2026-04-16T00:00:00Z the operator Template plus how Claude shortens the write-up without hiding the decisions that matter. Prompt caching changed our unit economics tag:nexcur.ai,2026-04-15:/blog/prompt-caching-unit-economics 2026-04-15T00:00:00Z 2026-04-15T00:00:00Z the operator When a 140 page handbook is generated against the same 20k token context forty times, the cache pays for the whole engagement. Break-even math with numbers. The llms.txt discipline tag:nexcur.ai,2026-04-14:/blog/llms-txt-discipline 2026-04-14T00:00:00Z 2026-04-14T00:00:00Z the operator Most llms.txt files in the wild are empty, stale, or marketing copy. The six-section structure and quarterly review discipline that makes the file useful. OWASP LLM Top 10 as audit checks tag:nexcur.ai,2026-04-12:/blog/owasp-llm-top-10-audit-checks 2026-04-12T00:00:00Z 2026-04-12T00:00:00Z the operator Two concrete audit checks per OWASP LLM Top 10 category. Runnable by a security lead who does not need to become an AI researcher. The operator-to-senior ratio that works tag:nexcur.ai,2026-04-11:/blog/operator-senior-ratio 2026-04-11T00:00:00Z 2026-04-11T00:00:00Z the operator The inverted pyramid of AI-native consulting delivery, and the failure mode to avoid. The handbook thesis, defended tag:nexcur.ai,2026-04-10:/blog/handbook-thesis-defended 2026-04-10T00:00:00Z 2026-04-10T00:00:00Z the operator Why we ship a long-form document at the end of every engagement, and why the slide deck is about to die as a consulting artifact. Why we stopped writing one-shot prompts tag:nexcur.ai,2026-04-08:/blog/stopped-writing-one-shot-prompts 2026-04-08T00:00:00Z 2026-04-08T00:00:00Z the operator Every deliverable runs through four stages now: draft, self-critique, human edit, eval. How we broke the pipeline apart. Retainer vs sprint: picking the right shape tag:nexcur.ai,2026-04-05:/blog/retainer-vs-sprint 2026-04-05T00:00:00Z 2026-04-05T00:00:00Z the operator The decision tree we use to pick between retainer and sprint at the top of every sales conversation. The pentest report is a literary form, and most of them are bad tag:nexcur.ai,2026-04-03:/blog/pentest-report-literary-form 2026-04-03T00:00:00Z 2026-04-03T00:00:00Z the operator A taxonomy of why inherited pentest reports fail both executives and engineers, and what Claude specifically changes about the form. We measured share of answer for 400 queries across 4 LLMs tag:nexcur.ai,2026-03-28:/blog/share-of-answer-400-queries 2026-03-28T00:00:00Z 2026-03-28T00:00:00Z the operator A structural audit of what actually gets cited in Claude, ChatGPT, Perplexity, and Gemini answers. Data, not theory. Includes the raw export. Billing hourly in 2026 is a moral hazard tag:nexcur.ai,2026-03-22:/blog/hourly-billing-moral-hazard 2026-03-22T00:00:00Z 2026-03-22T00:00:00Z the operator When an operator plus Claude can do a week of classical consulting work in a day, the hourly rate incentivizes slower delivery. Extended thinking is a research budget, not a default tag:nexcur.ai,2026-03-14:/blog/extended-thinking-research-budget 2026-03-14T00:00:00Z 2026-03-14T00:00:00Z the operator Extended thinking is expensive compute. Treat it like any research budget: spend it where the return is visible, leave it off everywhere else. The IAM attack path every Series A has tag:nexcur.ai,2026-03-07:/blog/iam-attack-path-series-a 2026-03-07T00:00:00Z 2026-03-07T00:00:00Z the operator The root-to-production path that shows up at 70 percent of Series A SaaS we audit, and the three links in the chain that close it. Citations are the new backlinks tag:nexcur.ai,2026-02-28:/blog/citations-new-backlinks 2026-02-28T00:00:00Z 2026-02-28T00:00:00Z the operator For fifteen years the link graph was the currency of SEO. In the answer-engine era, the currency is the citation. The economics are not the same. The consulting firm P&L that Claude rewrites tag:nexcur.ai,2026-02-21:/blog/consulting-firm-pnl 2026-02-21T00:00:00Z 2026-02-21T00:00:00Z the operator Every line of the P&L shifts when the model is a team member. Revenue, cost, and margin reconsidered. The templates that survive a model upgrade tag:nexcur.ai,2026-02-14:/blog/templates-survive-model-upgrade 2026-02-14T00:00:00Z 2026-02-14T00:00:00Z the operator The principles that kept 80 percent of our template library intact across six Claude versions, and the two reasons the other 20 percent broke. Claude is not a pentester. A pentester with Claude is. tag:nexcur.ai,2026-02-07:/blog/claude-is-not-a-pentester 2026-02-07T00:00:00Z 2026-02-07T00:00:00Z the operator The line between Claude-generated and human-validated findings, and why vendors who hide the line are selling marketing as security. Why we stopped chasing featured snippets tag:nexcur.ai,2026-01-31:/blog/stopped-chasing-featured-snippets 2026-01-31T00:00:00Z 2026-01-31T00:00:00Z the operator Position zero was the optimization target for a decade. For most categories we work in, chasing it is now a waste of budget.