iiinova — where your digital self keeps evolving

Nextcloud's restore passed on every push for three weeks. On a machine that had never run the stack: 500 on every request. The dump never carried the app's database account. Same-host restores test the archive. Not the recovery. heyvaldemar.com/restore-dril...

heyvaldemar.com

Restore Drills on a Clean Machine — Measured RTO for 47 Stacks, 19 Seconds to 7 Minutes | VALDEMAR

Restore drills that measure RTO on a machine that never ran the stack, 47 templates weekly. Every same-host restore passed. The first clean one answered 500.

000

Vladimir Mikhalev @valdemar.ai · 06/10/2026

34 of the 100 images I pin would have had their next AI review written against a repo that does not exist. It ruled anyway: NEEDS ATTENTION, on notes nobody read. Ask the verdict what it read. heyvaldemar.com/ai-reviews-r...

heyvaldemar.com

AI Reviews That Read Nothing — Thirty-Four Verdicts in a Hundred | VALDEMAR

A third of the images my fleet pins would have had their next AI review written against nothing. How to prove what an AI review read before anyone acts on it.

000

Vladimir Mikhalev @valdemar.ai · 03/10/2026

ALB listener from the AWS console: post-quantum TLS. From the API, CDK or Terraform: ELBSecurityPolicy-2016-08. Eight Terraform pipelines, no AWS account, 115 planted violations. All caught. If ssl_policy says nothing, it says 2016. builder.aws.com/content/37VV...

builder.aws.com

Testing Terraform for AWS With No AWS Account: 164 Assertions, 115 Planted Violations | AWS Builder Center

The console gives an ALB listener post-quantum TLS; Terraform gives it a 2016 policy. Testing Terraform for AWS with no account, and proving the tests can fail.

011

Vladimir Mikhalev @valdemar.ai · 27/09/2026

153 planted bugs across fifteen repos with green "verified on every push" badges. Four tests stayed green. They could never have gone red. A control nobody has seen fail is a rumour with a badge. heyvaldemar.com/mutation-tes...

heyvaldemar.com

Mutation Testing for Infrastructure Code — 153 Planted Bugs, Four Tests That Could Not Fail | VALDEMAR

Mutation testing for infrastructure code, done by hand across fifteen repositories. 153 planted bugs, and four green tests that could never have failed.

000

Vladimir Mikhalev @valdemar.ai · 15/09/2026

Seventeen hours green over a core dump. The check used pgrep -f, so it matched the shell running it. It found itself. It could not fail. Ask any check to go red before you trust it. heyvaldemar.com/docker-healt...

heyvaldemar.com

Docker Health Checks That Cannot Fail — Seventeen Hours Green | VALDEMAR

A Docker health check whose pattern matches the shell running it can never go red. How to find the checks in your fleet that never failed, and prove each one.

000

Vladimir Mikhalev @valdemar.ai · 24/08/2026

An agent invoked every five minutes runs 8,640 times in thirty days. A daily governance sync evidences 30 of them. 0.35% of runs with a record. Article 12 asks for a per-event record. A schedule is not one. heyvaldemar.com/watsonx-gove...

heyvaldemar.com

watsonx.governance Enforcement Tracking — Sampled Compliance and the Unevidenced Run | VALDEMAR

EU AI Act Article 12 wants a per-event record. Enforcement Tracking evidences thresholds on a schedule. What that gap costs, and the join key that closes it.

000

Vladimir Mikhalev @valdemar.ai · 07/08/2026

Proud to officially join the @cncf.io Ambassadors program! My focus will be on bridging the gap between maintainers and enterprise decision-makers, delivering deep-dive post-mortems, and building regulated-enterprise content. #CNCF #CloudNative

Dark, minimalist 3D rendering of an enterprise server rack in a concrete data center. Represents cloud-native infrastructure and Kubernetes architecture for a CNCF Ambassador announcement by Vladimir Mikhalev.

030

Vladimir Mikhalev @valdemar.ai · 31/07/2026

Three agents finish in parallel. Main takes one landing at a time. METR's RCT: devs ran 19% slower with AI and estimated 20% faster. Now run that error three ways at once. Writing code got parallel. Landing stayed serial. heyvaldemar.com/merge-debt-2...

heyvaldemar.com

Merge Debt: The 2026 Bill Your Parallel AI Agents Are Running Up | VALDEMAR

Three agents finish in parallel. Main takes one landing at a time. METR's perception gap, the rebase tax, and why Merge Debt is the number nobody plots.

000

Vladimir Mikhalev @valdemar.ai · 03/07/2026

The blast radius of an agentic Terraform apply isn't one workspace. It's every workspace the token can reach, times every team that flips the flag without recording why. Scope the token before you debate the flag.

000

Vladimir Mikhalev @valdemar.ai · 01/07/2026

The Terraform MCP server went GA June 11. One flag — ENABLE_TF_OPERATIONS — lets an AI agent run apply against a production workspace. The agent's code is clean. What's missing is the record of which named human approved the apply. That's the SOC 2 finding. heyvaldemar.com/terraform-mc...

heyvaldemar.com

Terraform MCP server GA: the Apply Gate your auditor will ask about | VALDEMAR

HashiCorp's Terraform MCP server is GA and IBM Bob can write production IaC. ENABLE_TF_OPERATIONS separates a safe assistant from autonomous apply.

010

Vladimir Mikhalev @valdemar.ai · 26/06/2026

Most flaky tests aren't a tool problem. They're coupled to implementation that changes every sprint. Test the user's flow, not component internals. Refactor freely, tests hold. Cypress Ambassador Spotlight: www.cypress.io/blog/cypress... #Cypress

010

Vladimir Mikhalev @valdemar.ai · 16/06/2026

For a decade we got paid for what we knew. AI just wrote that asset down to zero. Reproducible knowledge is free now. Judgment — context, accountability, the call you own in production — is the only thing that compounds. The data and the 5-move protocol: heyvaldemar.com/your-knowled...

A dark studio scene. A metallic bar dissolves into glowing red particles on the left, while a solid crystalline bar glows teal on the right, reflected on a wet black surface.

010

Vladimir Mikhalev @valdemar.ai · 15/06/2026

That makes perfect sense. You basically need a FinOps champion embedded in the process. Someone who speaks both "AWS invoice" and "sprint points". Otherwise, cost never makes it into the definition of done.

100

Vladimir Mikhalev @valdemar.ai · 13/06/2026

100%. The technical part (instrumentation) is usually a solved problem. The real bottleneck is cultural: getting engineers to actually care about that feedback loop before the invoice arrives. Have you seen teams bridge that gap successfully?

110

Vladimir Mikhalev @valdemar.ai · 13/06/2026

The velocity dashboard hides the bill. AI code is up, but copy/paste spiked & refactoring dropped. As a @gitkraken.com Ambassador, I track reality: thinning your senior bench now is a fatal misallocation of talent. The 2026 risk analysis: heyvaldemar.com/senior-engin... #DevOps #AI

A dark, monolithic cube hovering in a deep teal void, emitting frayed cyan energy from its upper edges while a faint crimson glow bleeds onto the surface beneath it.

061

Vladimir Mikhalev @valdemar.ai · 13/06/2026

Exactly. The guessing phase is where FinOps goes to die. When you replace assumptions with telemetry, you stop burning capital and start building a sustainable AI control plane. Data over intuition, always.

100

Vladimir Mikhalev @valdemar.ai · 13/06/2026

Exactly. Observability is the foundation of the maturity framework. Once you bridge the visibility gap, then you can start architecting for performance, cost, and security. Until then, you’re just guessing.

100

Vladimir Mikhalev @valdemar.ai · 12/06/2026

Spot on, Amber. The $34K Cloudflare incident I noted in the piece is the exact shape of that failure. Financial observability and code provenance are two sides of the same missing control plane. The cloud bill is just the loudest alarm.

000

Vladimir Mikhalev @valdemar.ai · 12/06/2026

You cannot model what you cannot measure. A shared maturity model is the end state, but architectural visibility is the absolute prerequisite. Right now, most organizations are failing step one. You don't build a governance model on top of shadow IT.

100

Vladimir Mikhalev @valdemar.ai · 12/06/2026

Your VP of Engineering doesn't know how many AI agents are coding in parallel today. I call this 2026 baseline Agent Sprawl. It’s an unmonitored liability. @gitkraken.com's new Agent Sessions View is the first real control plane for this. heyvaldemar.com/agent-sprawl... #AIGovernance #DevOps

A dark, metallic geometric cube floating in a dark environment, heavily entangled by chaotic, glowing cyan and red cables spreading outward. A visual representation of Agent Sprawl and unmonitored enterprise architecture.

161

Vladimir Mikhalev @valdemar.ai · 08/06/2026

Docker Captain, renewed for 2026. Three years in. Most container security tooling is built for the demo. The audit log doesn't care how the dashboard looks. Ship what survives review.

000

Vladimir Mikhalev @valdemar.ai · 26/05/2026

New on Docker's blog: I let Claude Code rewrite my entire blog — 146 posts, 6,024 images. It worked. I also stopped understanding my own codebase. That feeling is exactly why @docker.com had to build Sandboxes. www.docker.com/blog/untrust... #DockerSandboxes #AIAgents

docker.com

The Untrusted Autonomous Workload and AI Sandboxes | Docker

Learn why AI coding agents need stronger isolation, how Docker Sandboxes use microVMs, and what secure autonomous workloads require.

010

Vladimir Mikhalev @valdemar.ai · 25/05/2026

Renewed for year two as an AWS Community Builder. After editing a Packt book on container security last year, I keep coming back to one thing: most tooling in this space is theater. Audit logs don't lie. Dashboards do. Year two: less theater.

Official AWS Community Builder Year 2 badge: a large orange numeral 2 centered inside a hexagonal frame, encircled by a dark teal ring on a black background.

000

Vladimir Mikhalev @valdemar.ai · 19/05/2026

Yeah. Perception, memory, tools - pull one out and the whole "agent" thing collapses.

100

Vladimir Mikhalev @valdemar.ai · 18/05/2026

Works great until you ask it to actually do something with your tickets, repos, or cloud bills.

000

Vladimir Mikhalev @valdemar.ai · 23/04/2026

Docker Scout D → B. OpenSSF Scorecard 7.8/10. My aws-kubectl (700K+ pulls) now ships cosign via @docker.com + Sigstore, SPDX SBOM, SLSA provenance, non-root default, immutable tags. Framework + the 4-hour incident: heyvaldemar.com/docker-suppl... #Docker #SupplyChain #DevSecOps

Dark datacenter corridor in cinematic lighting with parallel black server racks on both sides, red LED status indicators reflecting on a wet concrete floor, overhead ceiling lights fading into mist, and a massive red hexagonal attestation glyph glowing on the right wall — symbolizing Docker supply chain hardening through cosign signatures and SLSA provenance.

020

Vladimir Mikhalev @valdemar.ai · 21/04/2026

That security verification issue in China is the part people outside the region never see documented. Thanks for the concrete field report - useful context for anyone deploying Cloudflare with a China audience. The honest tradeoff naming is what makes free services work long term.

000

Vladimir Mikhalev @valdemar.ai · 20/04/2026

Region matters a lot for Cloudflare latency. Free tier routes through nearest PoP, so distance to Cloudflare infrastructure changes the experience heavily. NA/EU is fast, other regions vary. Free is a fair trade if the latency works for your users.

100

Vladimir Mikhalev @valdemar.ai · 20/04/2026

Clarity is a different category - session replay and heatmaps, not just pageview analytics. Useful signal if you need to see how users actually move through the page. Heavier than a pageview beacon (~30 KB with replay on), but that cost buys real UX insight, not just traffic counts.

010

Vladimir Mikhalev @valdemar.ai · 20/04/2026

Fair correction. "Zero-runtime" was wrong - beacon.min.js is client-side, ~4-5 KB gzipped, just defer-loaded and not in the app bundle. The accurate framing is zero-SDK-in-bundle and zero-config, not zero-runtime. GA4 at ~70 KiB is still the distinction that matters.

100

Vladimir Mikhalev @valdemar.ai · 20/04/2026

That shift is real - MCP turns "do you have an API" into "can an agent discover, auth, and call it without human glue". Different bar. For content sites the first three scores are the ceiling, so 75/100 is a reasonable pass. API surface only matters if the product has one to expose.

010

Vladimir Mikhalev @valdemar.ai · 20/04/2026

Astro + Cloudflare Pages is a good pairing. Cut GA4 from heyvaldemar.com this week on the same stack, edge analytics made it a one-commit migration. Curious about that Level 4 breakdown - is that an MCP / agent-readiness audit tool? Haven’t seen that scoring surface before.

110

Vladimir Mikhalev @valdemar.ai · 20/04/2026

Too early to claim ranking impact. Only two weeks of data, and I cut GA4 the same day I added fetchpriority on the LCP image, so the signals aren’t isolated. Mechanism is real though: CrUX p75 LCP feeds CWV, CWV is a ranking input. Posting real-user numbers in 4-6 weeks once RUM fills in.

000

Vladimir Mikhalev @valdemar.ai · 20/04/2026

umami is solid. Architectural difference: umami still ships a ~2 KiB client script, Cloudflare injects beacon.min.js at the proxy so there is zero SDK in the bundle. Different tradeoff - umami wins on data portability, CF wins on zero-runtime. Both beat GA4 on a content site.

100

Vladimir Mikhalev @valdemar.ai · 19/04/2026

Removed Google Analytics 4 from the blog. Mobile Lighthouse 88 → 100, third-party cookies to zero. Cloudflare Web Analytics was already running at the proxy layer. GA4 on top was ~70 KiB of gtag.js paying for nothing. heyvaldemar.com/cloudflare-w... #WebPerformance #Cloudflare #Privacy

PageSpeed Insights mobile report showing performance scores before removing Google Analytics 4 from the Astro blog proxied through Cloudflare. Four score circles at the top: Performance 88 in orange, Accessibility 100, Best Practices 92, and SEO 100 in green. A large Performance gauge below displays 88 with a partially-filled orange ring.

PageSpeed Insights mobile report showing performance scores after removing Google Analytics 4 and relying only on Cloudflare Web Analytics injected at the proxy layer. Four score circles at the top, all in green: Performance 94, Accessibility 100, Best Practices 100, and SEO 100. A large Performance gauge below displays 94 with a nearly complete green ring.

461

Vladimir Mikhalev @valdemar.ai · 16/04/2026

Kill switch has to live outside the agent’s blast radius - separate identity, separate network path, credentials it can’t reach. Physical unplug means every other control already failed. That’s the Command Layer.

110

Vladimir Mikhalev @valdemar.ai · 13/04/2026

Both, but tool quality was the one I couldn't engineer around. Worktrees fix the stomping - they don't fix an agent acting on a tool response that's technically valid but semantically wrong. That's where I kept landing. What's your setup for catching that?

000

Vladimir Mikhalev @valdemar.ai · 13/04/2026

7M views on the "Mac Mini + AI agent, walk away" reel. The creator got hired by OpenAI. I bought the hardware and ran the thing against real work. Verdict isn't what the comments section thinks it is. youtube.com/shorts/YHyCn... #AIAgents #DevOps #AI #TechReview

A cinematic, high-contrast photograph of a silver Apple Mac Mini centered on a dark, highly polished boardroom table. The device and its crisp reflection are illuminated by a single overhead spotlight. In the softly blurred background, empty executive chairs line the table leading toward architectural glass panels bathed in a cold, diffused blue light. The atmosphere is quiet, corporate, and intensely focused on the hardware as a point of authority.

320

Reposted by Vladimir Mikhalev

Tatiana Mikhaleva @devops.pink · 13/04/2026

Want to let AI agents go full YOLO mode without risking your host? 🚀 See how @docker.com solves the isolation problem with Docker Sandboxes. Lightweight microVMs keep your system completely safe.🛡️ Watch here (and see who chewed my hat 🐶): youtu.be/PVjuMgjr0CU #AI #Docker

A smiling woman pulls a loose thread from the top of her blue Docker Captain beanie, showing where it was chewed. On the left side against a dark blue background, bold white text reads "DOCKER SANDBOXES" with the terminal command "rm -rf /hat" below it.

011

Vladimir Mikhalev @valdemar.ai · 18/03/2026

Competing with Devin AI for execution work is a race to the bottom. The only survival path for junior engineers is to become the Architect. You need a blueprint to make the jump. Watch the Field CTO verdict: youtube.com/shorts/xFgtz... #TechCareers #Architecture #AI #DevOps

A high-contrast, vertically split cinematic image of a dark tech corporate boardroom. The left half displays chaotic, tangled red data streams over a dark conference table, symbolizing the stressful career trap of execution-level tech labor. The right half reveals a clean, structured white architectural blueprint illuminating a solitary premium black leather executive chair, representing Field CTO leadership, governance, and control. Neo-noir aesthetic, illustrating the tech industry's stark divide between competing with AI execution and commanding software architecture.

011

Vladimir Mikhalev @valdemar.ai · 16/03/2026

300M private AI messages exposed. Not a hack—a default Firebase misconfiguration found in >50% of AI apps. I built the 3-layer Operational Discipline Protocol to shut down AI sprawl before regulators do. The Verdict: heyvaldemar.com/exposed-ai-m... #DevSecOps #CISO #AI

Visual metaphor for a massive AI data breach and cloud misconfiguration. A monolithic dark database cube is crumbling, with millions of sensitive user messages (glowing blue data particles) leaking uncontrollably into a dark, unprotected void. A stark red alert light emanates from the base, symbolizing imminent risk and CISO accountability. AI security and data loss prevention concept.

020

Vladimir Mikhalev @valdemar.ai · 13/03/2026

Execution lives in the terminal. Architecture requires the graph. Cognitive load is a liability. You don’t need more terminal aliases - you need absolute visibility over your blast radius. @devops.pink breaking down the visual control plane we mandate.

000

Vladimir Mikhalev @valdemar.ai · 13/03/2026

Visibility is control. The terminal executes, but the graph governs. Excellent breakdown of the standard.

000

Vladimir Mikhalev @valdemar.ai · 13/03/2026

Devin AI is filtering out everyone who thought "being technical" was a career. The market is separating labor from leadership. Don't get caught in the execution trench. Watch the Field CTO verdict: youtube.com/shorts/LlvWZ... #TechCareers #TechLeadership #DevOps #AI #FutureOfWork

A cinematic, high-end corporate boardroom designed in a neo-noir dark tech aesthetic. A massive circular opening in the ceiling funnels a chaotic, glowing red stream of complex data down onto a sleek, round conference table. The table is surrounded by empty black executive chairs, symbolizing the filtering out of execution-level labor. The surrounding brutalist room is bathed in cold, muted cyan light, creating deep shadows and an atmosphere of institutional dread and strict architectural control. Kojima style, quiet luxury, executive leadership.

000

Vladimir Mikhalev @valdemar.ai · 12/03/2026

The copy-paste lifecycle is a race to the bottom, Patrick. The market no longer pays for hands on a keyboard; it pays for the verdict and risk ownership. Glad you see the board clearly.

100

Vladimir Mikhalev @valdemar.ai · 12/03/2026

Stop competing with AI on execution. The only survival mechanism is the Field CTO class. Own the decision, set the mandate, and command the agents. Watch the final verdict here: youtube.com/shorts/G8eQY... #TechLeadership #Architecture #AI #DevOps

A dramatic high-angle overhead view of a massive brutalist concrete corporate command center. A solitary male executive silhouette stands at the far head of an immense concrete conference table. The top surface of the table is a luminous projection, displaying a dynamic, glowing matrix of white global cloud formations and data patterns, resembling a strategic atmospheric data map. Two perfect rows of empty, dark leather executive chairs align the length of the table, emphasizing the Principal's isolation and architectural scale. Powerful beams of cold light cut through the atmospheric haze from the top left, illuminating the glowing map. The surrounding space is built with heavy brutalist concrete columns, and massive windows offer a glimpse of a sprawling, cold cyberpunk metropolis fading into darkness. A single, striking pool of vivid red reflection burns on the dark polished concrete floor in the foreground, contrasting sharply with the cold blue data map. Neo-noir, Dark Tech, institutional command perspective.

100

Vladimir Mikhalev @valdemar.ai · 11/03/2026

Pivoting to "AI Engineer" is a career trap. You aren't escaping automation; you're running straight into a crowded execution market. Stop competing on tasks. Learn how to control the blast radius. Watch the Field CTO verdict: youtube.com/shorts/jhzie... #TechCareers #AI #Architecture

A cinematic, dimly lit corporate boardroom designed in a dark tech, neo-noir aesthetic. At the center, a long conference table features a rigid, glowing red architectural grid embedded in its glass surface, flanked by perfectly aligned, empty black leather executive chairs. In the background, a massive wall-sized screen displays a cold, cyan-tinted futuristic cityscape. The highly polished dark floor reflects the stark contrast between the structured red governance of the table and the sprawling blue complexity of the city. Quiet luxury, institutional dread, and high-end executive control.

000

Vladimir Mikhalev @valdemar.ai · 10/03/2026

Context management is a huge part of it, yes. My point is less “local doesn’t work” and more “the Instagram version hides the control layer.” Once you add tools, memory, retries, permissions, and edge cases, the blast radius shows up fast. Appreciate the write-up.

110

Vladimir Mikhalev @valdemar.ai · 09/03/2026

Cognition Labs and Infosys just made execution-level engineering obsolete. Competing with Devin AI on tasks is a trap. The only leverage left in this market is architectural control and risk governance. Watch the full Field CTO breakdown here: youtube.com/shorts/FJc7v... #Tech #AI #DevOps

000