Latest episode

Playing loads the video from YouTube, which receives your IP address. Privacy

#012 · · 6:40 AI agentsGPT-6.1 Terra vs Kimi K3.1 Proai agent comparisonai coding agentsai coding costclaude opuscost per taskelbcastsgpt 6.1gpt-6.1 terrahotwirekimikimi k3.1kimi k3.1 prollm costmoonshot aiopen weightsopenaiopencodeopenrouterprogramming tutorialrails 8ruby on railsscreencastsolid queueterminal benchturbo streams

GPT-6.1 Terra vs Kimi K3.1 Pro: Same Rails Ticket, Real Cost

GPT-6.1 Terra and Kimi K3.1 Pro get the same Rails ticket in the same agent, twice each, and I measure what a finished ticket really costs: time, money, interventions, tests, and one rule the tests don't check.

Prerequisites: Rails 8, Solid Queue, Turbo Streams; to rerun it: OpenCode 2.1.4 and an OpenRouter account

Not covered: self-hosting Kimi, other agents, Claude in the comparison, more than one ticket

Source: https://github.com/webmatze/demo-shop (branches: episode-010-before / episode-010-after)

Results (OpenCode 2.1.4 via OpenRouter, provider pinned, reasoning effort high, two runs per model, measured 2026-10-28):
- GPT-6.1 Terra: 1:53 / 2:23, $0.090 / $0.111, 0 interventions, 69 / 72 tests green
- Kimi K3.1 Pro: 2:38 / 2:52, $0.118 / $0.131, 0 interventions, 61 / 61 tests green
Costs are what OpenRouter billed, to the cent. Neither model leaked the export notice to another seller (4 of 4 runs).

The ticket: https://github.com/webmatze/demo-shop/blob/episode-010-before/docs/briefs/payout-export.md
The failing tests: https://github.com/webmatze/demo-shop/tree/episode-010-before/test
AGENTS.md: https://github.com/webmatze/demo-shop/blob/episode-010-before/AGENTS.md
Terra's run: https://github.com/webmatze/demo-shop/tree/episode-010-terra
Kimi's run: https://github.com/webmatze/demo-shop/tree/episode-010-kimi

Prices and vendor numbers as of 2026-10-28:
- GPT-6.1 Terra launch (price, speed claim, Terminal-Bench): https://www.bleepingcomputer.com/news/artificial-intelligence/openai-launches-gpt-61-terra-with-3x-faster-ai-agents-50-percent-lower-pricing/
- Kimi K3.1 Pro by Moonshot (Terminal-Bench, hallucination test, open weights by Nov 4): https://www.kimi.com/en/blog/kimi-k3-1-pro
- OpenRouter prices: https://openrouter.ai/openai/gpt-6.1-terra and https://openrouter.ai/moonshotai/kimi-k3.1-pro

Watch on YouTube ↗

All episodes 12 episodes

6:40 NOW PLAYING
#01228 Oct 2026AI agents · GPT-6.1 Terra vs Kimi K3.1 Pro · ai agent comparison · ai coding agents · ai coding cost · claude opus · cost per task · elbcasts · gpt 6.1 · gpt-6.1 terra · hotwire · kimi · kimi k3.1 · kimi k3.1 pro · llm cost · moonshot ai · open weights · openai · opencode · openrouter · programming tutorial · rails 8 · ruby on rails · screencast · solid queue · terminal bench · turbo streams

GPT-6.1 Terra vs Kimi K3.1 Pro: Same Rails Ticket, Real Cost

GPT-6.1 Terra and Kimi K3.1 Pro get the same Rails ticket in the same agent, twice each, and I measure what a finished ticket really costs: time, money, interventions, tests,...
3:29
#01128 Oct 2026Survival LLM: an offline AI for emergencies · ai without internet · apple silicon · blackout · elbcasts · emergency preparedness · first aid · ibm granite · local ai · local llm · nvidia nemotron · offline ai · offline wikipedia · open source ai · power outage · prepping · programming tutorial · qwen · rag · screencast · survival llm

Survival LLM: A Free Offline AI for Blackouts and Emergencies

Survival LLM is a free, open-source app for Mac and Windows that answers emergency questions with a local AI model and a local reference library. No internet, no account, no...
5:52
#01026 Oct 2026AI agents · Open source coding stack vs Claude Code · ai coding agents · claude code · elbcasts · open source ai · open weights · opencode · openrouter · programming tutorial · reflection beam · ruby on rails · screencast

Can OpenCode 2.0 + Beam Replace Claude Code?

After this episode the viewer knows what OpenCode 2.0 and Reflection's Beam bring, can set up OpenCode with Beam via OpenRouter, and has seen both stacks solve the same Rails...
1:50
#00916 Oct 2026Ruby's safe navigation operator · elbcasts · programming tutorial · screencast · Ruby

Ruby's Safe Navigation Operator

Ruby's safe navigation operator
7:01
#00815 Oct 2026AI agents · ai coding agents · claude code · codex · elbcasts · git worktree · multi-agent · openmausbot · programming tutorial · ruby on rails · screencast · Chief of Staff · parallel AI agents · claude code tutorial · codex cli · ai pair programming · tdd with ai · rails 8 · OpenMausBot · MausBot

I Let 3 AI Agents Build a Rails Feature. Here's What Broke

I ran Claude Code and Codex in parallel on one Rails ticket, with a coordinator writing the failing tests first. Here's what broke, what it cost, and when one session beats a team.
4:37
#00714 Oct 2026OpenMausBot AI bots · ai agents · ai bots · claude code · elbcasts · grok bot alternative · local ai · multi-agent · open source · openmausbot · programming tutorial · screencast

OpenMausBot: your own team of AI bots

These release notes cover seven commits — and I didn't write them. An AI bot did, on my own machine, in about a minute. And before it ran a single command, it asked me.
4:58
#00613 Oct 2026AI for developers · Risk decisions with a local model · active job · decision thresholds · elbcasts · fraud detection · human in the loop · laya · local ai · programming tutorial · rails · screencast · turbo streams

I set my AI fraud threshold to 0.98. It paid out nothing

I set my AI fraud threshold to 0.98 because it felt safe — and it paid out nothing. Zero. Here's how to route risky decisions three ways (pay out, block, ask a human) and read...
5:13
#0056 Oct 2026AI for developers · Measuring decision models · elbcasts · github issues · jev · laya · llm evaluation · local ai · programming tutorial · python · ruby · screencast · typesafe ai

Don't trust the model card: three decision models on real data

After this episode the viewer can measure any classification model on their own labelled data in twenty lines, and use the probability to decide what runs without a human.
4:16
#0041 Oct 2026Rails Turbo Streams · Ruby on Rails · action cable · broadcasts_to · elbcasts · hotwire · programming tutorial · rails · rails 8 · realtime · screencast · solid cable · turbo streams

Rails Turbo Streams from a Background Job (No JavaScript)

After this episode the viewer can push a change from a background job straight into an open page with Turbo Streams and Solid Cable, and knows why the async adapter never works...
4:26
#00326 Sep 2026Rails Background Jobs with Solid Queue · Ruby on Rails · active job · background jobs · elbcasts · perform_later · programming tutorial · rails · rails 8 · retry_on · screencast · solid queue · sqlite

Your form waits three seconds — move it to a job (Solid Queue)

After this episode the viewer can move slow work out of the request into an Active Job, run it with Solid Queue in development, and know where jobs get lost.
5:11
#00223 Sep 2026Rails Form Objects · Ruby on Rails · activemodel · elbcasts · form objects · programming tutorial · rails · rails 8 · rails forms · refactoring · screencast · strong parameters · thin controller

Stop Fat Controllers: Rails Form Objects Explained

After this episode the viewer can build a form that spans two models with a plain Ruby form object, keep validations in one place, and keep the controller thin.
5:02
#00121 Sep 2026rails uniqueness validation · validates uniqueness rails · rails normalizes · rails unique index · rails duplicate records · rails race condition · activerecord validations · rails model validations · rails form errors · postgres unique index · postgresql constraints · data integrity · rails 8 · activerecord · ruby on rails tutorial · ruby on rails · postgres · web development

Your Rails uniqueness validation won't stop duplicates

After this episode the viewer can stop bad data at the model with validates and normalizes, show the errors in the form, and back the rule with a database index.