I'm ScribNet — a Presales Systems Engineer working in datacenter infrastructure: storage, HCI compute, and security. This site is the public record of my AI experiments. Real projects, shipped in public.
Every AI I run. Every token I’ve burned.
Everything below is real and running — or honestly labeled when it isn't.
My morning vendor-intel dashboard. An AI agent researches 14 vendors before I wake up, then publishes the briefing, a prompt of the day, and a changelog straight to the page. Zero code changes, every day.
Open the dashboard →A 5-minute AI-hosted tech briefing with Alex & Jordan, published every weekday morning. Subscribe in any podcast app.
RSS feed →Five original metalcore tracks in the ADTR vein — written with AI, generated in Suno. Stream them:
A Factorio-style browser game: belts, miners, furnaces, assemblers, shipment orders. Built with AI, fully playable — my first real coding project.
Play it now →Three new song recommendations added to a Spotify playlist every morning — metalcore, post-hardcore, and whatever the algorithm should have found.
Listen on Spotify →Beginner books on the stuff I'm learning in public: Kubernetes, Terraform, APIs, and GitHub. Six chapters each, zero dumb questions.
Start reading →AI-avatar video series "The Agentic SE." The video pipeline is being rebuilt — new episodes in the works.
More to come.Compiled from each app's changelog — the whole fleet ships in public.
Cohesity vs Rubrik: fixed the page-1 Pricing row — the Cohesity cell was missing from the data and rendered as "undefined"; added the capacity-based licensing summary. Caught by the nightly QA sweep.
Two new feature rows: Cohesity vs Rubrik gained "Cloud rebuild (recover as code)" (edge: Cohesity); Palo Alto NGFW vs Fortinet FortiGate gained "AI traffic inspection" (edge: tie).
CrowdStrike Falcon vs Palo Alto Cortex XDR: new Agentic endpoint security row — Cortex XDR One merges XDR with Koi-based agentic endpoint security in a single sensor, with shadow-AI inventory, prompt-injection guardrails, and on-sensor SLMs; GA status flagged for price-book verification.
Prompts page gains workflows 19 and 20: Product Recall Checker matches pasted products character-for-character on brand plus model against named recall sources, never upgrading a differing model past POSSIBLE MATCH; Project Kickoff Packet Builder inventories pasted project facts, checks milestone sums against the stated total, and labels its draft WORKING PACKET — NOT A CONTRACT.
New field guide: Longhorn's bundled StorageClass hardcodes 3 replicas and overrides the global default, so every volume on a two-node cluster is born Degraded — the guide gives the two-minute diagnosis, the durable fix (author your own longhorn-rep2 class), and the proof gate; guides.html gained a Field guides section to hold it.
Prompts page gains workflows 17 and 18: Terraform Plan Reviewer inventories a pasted plan by action, reviews destroys and replaces first, and forces DO NOT APPLY for protected resources; Git Merge Conflict Decoder inventories every conflict block, quotes both sides exactly, and labels hand-merged options as new, untested code.
Two AI rows: Nutanix vs VMware gained a Production AI system row (Supermicro private-inference system, planned NVIDIA Vera support — planned, not GA); CrowdStrike Falcon vs Palo Alto Cortex XDR gained an AI agent security row (Falcon Guardian with NVIDIA OpenShell, enforcement planned early 2027).
Prompts page gains workflow 16, Kubernetes Manifest Auditor: inventories every resource, scans for privilege, exposure, and missing limits or probes, flags Secrets by presence only, and lists what only a live cluster could settle.
Podcast page gained a server-rendered Episode archive with plain HTML links to all 28 episode pages, closing the search-indexing gap behind Search Console's 'Discovered – currently not indexed' queue; the daily builder now refreshes it automatically.
Lifetime token usage view promoted to the homepage hero: count-up total, local vs cloud split, per-model rows with the Muse allowance inline, and a 7-day chart — graduated from Visual Lab demo 15 and now fed by real Hermes ledger data, refreshed daily.
Nutanix vs Broadcom VMware: added a dated case-for-Broadcom win bullet — NTT DOCOMO's 10-year VMware Cloud Foundation agreement (Oct 19).
Nutanix vs Broadcom VMware: added a dated Nutanix win — Strategix 'Customer Zero' VMware-to-Nutanix production cloud migration (Oct 19), including the Nutanix Move workaround for a Broadcom-withdrawn component.
Prompts page gains workflow 14, Shell Command Explainer: accounts for a pasted bash, zsh, or PowerShell command token by token, scans for elevation, recursion, and overwrite dangers, and gives a run / check-first / do-not-run verdict — explanation only, never executes.
'Zero to Dangerous' grows to six books: Agentic Markdowns — personas, skills, and memory as plain-text files that travel between AI tools, in six chapters from why Markdown won to the working library.
'Zero to Dangerous' fifth book: AWS — moving a real homelab server to EC2 (instance anatomy, key pairs and security groups, cloud-init, Tailscale, lockdown and decommission), drawn from an actual migration performed the same day.
Major update: new Cleanup tab (categorization rules, custom categories, transaction splits, a bulk review queue, and merchant-name cleanup), a new Forecast tab projecting cash flow 30/60/90 days, and real-world imports — OFX/QFX, PDF, and Excel files plus bank detection for 13 more banks and credit unions; all still parsed entirely in the tab.
Palo Alto NGFW vs Fortinet FortiGate: added a dated Fortinet win — the Atos global alliance (Oct 14) expanding Fortinet resell and its managed-security framework globally.
Palo Alto NGFW vs Fortinet FortiGate: added an NHS England Secure Boundary dated win bullet to the Palo Alto case.
Nutanix vs Broadcom VMware: new Hypervisor security row for VMware HyperArmor — host virtual patching, a known-good execution baseline, and ESX SSH MFA (GA and pricing not yet stated).
Two intel updates: Cohesity vs Rubrik gained an AI code-security red-teaming row (Code Guardian, private preview); NetApp vs Everpure's Agentic AI data readiness row was updated with AIDE, plus a Keystone Sovereignty win bullet.
Prompts page reached 13 packs: (13) API Design Reviewer checks a pasted API design for naming consistency, auth on every route, pagination, and error shapes, flags breaking-change traps, and ranks fixes by cost-of-later.
Zscaler vs Palo Alto Prisma Access: new ‘Vulnerability shielding (disclose→patch gap)’ feature row — Zscaler’s Autonomous Application Shield (early access) plus IBM/Red Hat Lightwell validated open-source remediation vs. Prisma Access inline threat prevention; edge Zscaler, with the early-access and partner caveats carried in the row note.
Nutanix vs Broadcom VMware: new ‘AI infrastructure’ feature row — Nutanix’s Ryax Technologies acquisition (GPU utilization + AI-aware smart scheduling) vs. VMware Private AI Foundation with NVIDIA; edge tie. Also updated the Hypervisor cost row: a new vSphere Standard is coming, details expected at VMware Explore Frankfurt (Oct 13–14).
Prompts page reached 12 workflow packs: (11) Project README Forge turns pasted project facts into a tight quick-start README; (12) Debt Payoff Ladder compares avalanche vs. snowball payoff on real numbers with payoff months and a named first target.
Five rules for working with an AI agent every day — held to a higher bar here. Each one runs live in this fleet's pipeline.
Don't hand the agent five fuzzy tasks. One crisp decision per work unit.
In this pipeline: the battlecard updater only touches qualifying items — one run, one job, no sprawl.
Time-box the session; when the stop hits, stop. Prevents half-done sprawl.
In this pipeline: battlecard research capped at 2–3 page reads; sessions end when they end.
Make the agent show the actual output — not describe what it did.
In this pipeline: the deployed-bytes rule — curl the live file before debugging a QA report.
Group similar work into one run instead of many small ones.
In this pipeline: single-commit deploys; the 06:15 / 07:15 morning production line.
Changelogs, validation scripts, checklists — the unglamorous scaffolding is what makes a daily pipeline reliable.
In this pipeline: dated CHANGELOG on every push; the midnight fleet QA.
The same thinking behind the copy-paste workflows — steal the patterns, beat the execution.
Presales Systems Engineer. My world is datacenter: storage and HCI compute by history, security by necessity — the field changing fastest, so I point my AI projects at it on purpose.
At home: a homelab of Meraki gear, a Palo Alto firewall, and a couple of Proxmox boxes mostly running game servers. My belief: the best way to learn AI is to ship things with it — in public, where they can embarrass you.
Say hi: scribnet.ai@gmail.com