Anton Babenko 🇺🇦

@compliancetf / @weeklytf / / AWS Hero / Terraform influencer 🇺🇦🇳🇴

[Anton Babenko 🇺🇦 @antonbabenko]

While working on @compliancetf, I created a Claude skill called terraform-skill (terraform-best-practices + other trusted sources).

In 2026, please stop saying "AI hallucinates with #Terraform". Use HashiCorp MCP + this skill so answers are grounded in real docs and battle-tested patterns.

Feedback welcome, and shares appreciated. 🚀

https://t.co/XtS23Q63FD

[Alexander Whedon @alex_whedon]

Introducing SubQ Snap, the first long-context decision model for real enterprise work.

Snap is the only decision model or LLM where compute per token stays constant as context grows.

• 384K tokens of multimodal context, 12x larger than Jev.
• 61.7 on Decision Index v0.2.1 vs Jev’s 57.9, outperforming on 13 of 16 enterprise benchmarks.
•Processes 100K tokens and returns a decision in under one second.
•$0.05/MM input tokens. Output is free.

Available today to anyone.

The first 10,000 sign-ups get 100M free input tokens.

We’ll likely hit capacity fast and will add users as we add compute. Compute offers welcome 🙂.

Latency will be variable as we tune the platform to this new model and category of users.

Model speeds are 80ms for minimal context, 250ms for 32k, 660ms for 100k, and 2.5 seconds for 384k.

Visit: https://t.co/Nc76iJF2a8

[Anton Babenko 🇺🇦 @antonbabenko]

A follow-up to "Meet Deliberation": what changed between v3.5.2 and v3.28.0, and why the numbers only started changing my config once I could see each run. https://t.co/2SvygiskRK

[Anton Babenko 🇺🇦 @antonbabenko]

#Deliberation now has a local dashboard: every multi-model run drawn live, one lane per model, plus an analyzer that tells you which model agrees more than it adds.

https://t.co/dJpA57bGAB

#consensus #multimodel

[Anton Babenko 🇺🇦 @antonbabenko]

I will be presenting at AWS re:invent (COM302):
The judgment gap: when Terraform scans clean but fails the audit.

Monday @ Caesars Forum | Level 1 | Academy 414
13:30 - 14:30

More info in the official catalog: https://t.co/T4YIfO36AR

#ComplianceAsCode #Terraform #AWSreinvent #DevSecOps @compliancetf

[Anton Babenko 🇺🇦 @antonbabenko]

* majority of the terraform-aws-modules FORKS exists… (early morning for me) :)

[Anton Babenko 🇺🇦 @antonbabenko]

Majority of the terraform-aws-modules exists due to Terraform workarounds (e.g. lack of vars in lifecycle block).

Read my post on how you can simplify your usage by just swapping the modules source url.

This is @compliancetf and it is available for free!

https://t.co/AJU81Byh3g

[Compliance.tf @compliancetf]

😎 Six months ago we shipped Operational Rules for terraform-aws-modules. The question we get most since: "what exactly will it do to my module?"

Now you can see the diff before you touch a source line. Rules Playground and six more rules.

https://t.co/7FR3A1SRdt

[Anton Babenko 🇺🇦 @antonbabenko]

Since March, the most common question about Operational Rules: "show me the diff first".

Rules Playground is live. Pick a terraform-aws-modules module, tick rules, and read the diff. No account needed.

Also: 15 rules now, and org-wide Baselines on trial or paid plans.

https://t.co/TqowtgWQlc

#Terraform #ComplianceAsCode #DevOps

[Anton Babenko 🇺🇦 @antonbabenko]

Back from #AWSHeroesSummit where it was awesome to meet old and new friends from all over the world and have some fun.

My travel story ( https://t.co/YUvWpwUXWV ): 42 countries, 422 flights.

More to come...

[Compliance.tf @compliancetf]

tfpolicy (by HashiCorp) ships a policy stage that runs before Terraform downloads a module.

At that point a policy sees a source string and a version. It allows or denies a name.

Which module that name resolves to is doing a lot of the compliance work.

https://t.co/Oai5o8FtPa

[Tom Dörr @tom_doerr]

Terraform best practices for AI coding agents

https://t.co/I7MP5c6iQE

[Anton Babenko 🇺🇦 @antonbabenko]

Another open-source tool from building @compliancetf

repro-lambda builds AWS Lambda zips outside Terraform: byte-identical on macOS and CI, uploaded to S3 by content hash, then promoted dev -> prod with no rebuild. Terraform reads s3_existing_package instead of rebuilding on every plan.

#Terraform #AWS #Lambda #serverless

https://t.co/hanuSZjTHR

[Anton Babenko 🇺🇦 @antonbabenko]

no tie-break by design. Convergence is unanimous-APPROVE among responders + zero open criticals + arbiter can't self-approve. A disagreement isn't voted on - it triggers a revision round (arbiter must address each critical, peers re-review), capped at 5. Still split at the cap? It returns non-converged instead of picking a winner.

[Anton Babenko 🇺🇦 @antonbabenko]

@Validate_QA @OpenRouter not cut - meant "another peer's opinion." Each voice reviews in isolation, no panelist sees another's verdict. Zero cross-contamination, same way the arbiter is blind to the panel until it commits.

[Anton Babenko 🇺🇦 @antonbabenko]

Slightly different mechanism. The criteria aren't authored per-run, so there's nothing to leak - it's a fixed rubric (correctness, security, scope, ambiguity, perf, ops), identical for every voice. What pre-commits is the arbiter's own verdict, recorded before the panel is revealed. Peers vote blind too, none sees another's. And the arbiter can't self-approve: convergence needs the external voices to carry APPROVE, not just the arbiter.

[Anton Babenko 🇺🇦 @antonbabenko]

@Validate_QA @OpenRouter Exactly, blind pre-commit input by the arbiter is already implemented in Deliberation.

[Anton Babenko 🇺🇦 @antonbabenko]

"Panels of models which consistently outperforms individual models" - this is exactly the point I was chasing when working on the Deliberation.

https://t.co/5WpSEm1Mrq

Ask Codex, Gemini, Grok, and 400+ OpenRouter models (Qwen, Kimi, DeepSeek) for second opinions or arbiter-mediated consensus. One MCP server for Claude Code, Codex, Cursor, Kiro, OpenCode. Measures which models earn their seat.

#openrouter #fusion

[OpenRouter @OpenRouter]

Introducing the Fusion API, the smartest compound model in the market.

Fusion achieves Fable-level intelligence at half the price.

How it works 👇

[Anton Babenko 🇺🇦 @antonbabenko]

@iamlukasniessen @compliancetf Hi Lukas! It's not the case for quiet a while already. Feel free to review it again and update the README in your repository to reflect the reality.

[Anton Babenko 🇺🇦 @antonbabenko]

While working on @compliancetf, I created a Claude skill called terraform-skill (terraform-best-practices + other trusted sources).

In 2026, please stop saying "AI hallucinates with #Terraform". Use HashiCorp MCP + this skill so answers are grounded in real docs and battle-tested patterns.

Feedback welcome, and shares appreciated. 🚀

https://t.co/XtS23Q63FD