Ariel and Igal Talk Bresleveloper vs SlimSoft — and How AI Helps Lawyers

One long conversation. Two veteran developers. Two business models. Ariel (Bresleveloper) versus Igal (SlimSoft) — talking models, code, and money. And in the middle, with no warning: a live demo of an AI tool for lawyers, with real documents.

The clip to watch first

Around the 14:15 mark of the full conversation, Ariel stops talking, opens a screen, and shows an AI app for lawyers. The idea in one line: a lawyer uploads their cases and documents, and the system helps analyze them, find connections, and draft outputs — instead of hours of manual work.

What do you see there?

• A dashboard for a lawyer — documents, cases, medical records, and client correspondence, all in one screen.
• A million medical documents? Sorted automatically — upload them and the system organizes them by topic.
• ~30 documents, clause-by-clause analysis — the system reads the case, goes through the contract clause by clause, calculates the strength of the causal link (who caused what), and estimates whether it is even worth suing the insurance company.
• Sources, not hallucination — the conclusions are backed by real case law pulled from professional databases, not “I think so.”
• A Word draft to download — all the analysis ends in a first-draft Word file the lawyer can open and continue working on.

This is not “ChatGPT open in a tab.” This is a tool tailored to a lawyer’s workflow, with real documents. Very real: during the demo they had to hide actual client names, and at some point someone asks, “Are you recording this?” Answer: “Obviously.”

The full conversation: Bresleveloper vs SlimSoft

The clip above is just one moment from a conversation of over half an hour between Ariel Rubinstein (Bresleveloper) and Igal Yerushalmi (SlimSoft). Two veteran developers, two business approaches, and open arguments about AI, code, models, and how to actually build a business in this space.

What came up in the conversation

A live conversation is not a slide deck — topics jump. Before you dive into the full video, here is what was actually in it, sorted by subject:

• Fable vs Opus series 4 — the conversation opens on what was new at the time: the Fable model and Claude Opus series 4. In benchmarks Opus looks stronger, but the claim in the conversation is that when you look at the actual code, Fable writes more elegantly, “knows what it is doing,” and solved a stubborn bug that Opus series 4 simply could not kill. On pricing: Opus 5.5 actually dropped in price compared to the previous version — but when you are on a subscription, nobody really cares.
• AFK work and splitting agents — one of them likes to let go and let the CLI run alone (permission mode); the other prefers to stay “human in the loop.” The practical advice for anyone frustrated: do not throw frontend and backend into the same agent — split them into two agents, each with a clean context, and suddenly it works. Most AI frustration is actually a management problem, not a model problem.
• Claude Code in the cloud, and CI/CD that feels good — running in Claude Code’s cloud environment with ~$100 of free credit (one project “burned” about $12). The really impressive part: a full pipeline — backend on a VPS, frontend on Firebase, merge to master and within a minute you have a live environment. And the cherry on top: the agent logs into the server logs on its own, diagnoses the issue, and fixes it. (There is a side joke about DevOps people becoming unemployed.)
• Zig — and the limit of AI — a discussion about Zig, the language that finally really competes with C. In code close to the kernel, you need to keep every pointer and every memory call precise in your head — exactly the context AI still cannot absorb. In high-level languages (Ruby, for example) everything is wrappers on wrappers, and there AI is already the main player, especially when you can give it a million tests to run.
• Experiment: Tower Defense from a single prompt — they gave many models one prompt: “Build me a Tower Defense in HTML/CSS/JS, no questions, you have permissions, go.” Result: only one model produced a game you could call “not bad at all,” Sonnet produced something that worked but weaker, and the rest failed. And the most depressing part? Every model wrote the exact same story — light versus darkness, candles and demons. Zero creativity, and that is exactly where the human re-enters the equation.
• The harness that turns a conversation into a product — one of them is expanding a harness that interviews you — about 30 questions — pulls out what you really want to build, and turns it into a product, all in the CLI (maybe a small UI in JavaScript later). The vision: a person with no coding knowledge enters, talks, and gets a result. Will be released when ready.
• And in the middle — the AI lawyer tool demo — exactly what you saw in the clip above: a dashboard with medical records, documents, and correspondence; a million medical documents sorting themselves; a case of ~30 documents analyzed clause by clause, including causal-link strength and lawsuit viability against an insurance company — all backed by real case law — ending in a downloadable Word draft. The sentence that explains it all: “This takes her hours.” And yes, at some point someone remembered the names on the screen were real.
• Bonus demo: free Design Pack — a tool that takes a website you like and returns a full design package: colors, fonts, layout, images, do’s and don’ts, even a prompt guide for agents. “I want my site to look like X” — and there is your design. Free, by the way.
• AI influencers — people running AI celebrities, selling workbooks on Stan, making about $2,000 a month “without working.” The theory in the conversation: content does not need to be real — money needs to be real.
• Money, restaurants, and the Shmandrix dream — a short dive into business: thin margins in restaurants, a famous chef who closed his special restaurant (₪300–350 per dish, month-long waiting list) because it was not profitable — in restaurants only chains really make money. Along the way a concept is born: a food chain in the style of Cofix, but with portions that actually fill you up — a 3kg schnitzel, a ₪150 burger, and an opening catering event with influencers and 50 insane recipes.
• Omarchy — the big crush — later in the conversation the operating system Omarchy comes up: a free Linux distribution based on Arch, from the person behind Ruby on Rails (who now writes code with only AI, and told the Rails conference audience: if you are still writing code manually without AI, you are a loser). It comes with everything pre-installed, so many keyboard shortcuts that “you can live without a mouse,” and has an agent skill: “Install VS for me” — and a moment later you have Visual Studio (they demoed installing Zoom live). Official statement: “I have never said about an OS that I am in love — I am in love.”
• The next computer — Mac, probably. After ₪8,000 of desktop with a graphics card that did not deliver the goods — no more desktops, a laptop is the solution. Also screens: dropping from four to two, and the next one is one of those monsters that splits into six.
• Windows standing on two legs — Office and gaming. Office is already “getting along” on Linux, and in gaming: Valve did huge work to make Steam games run well on Linux, and is releasing SteamOS based on Linux for computers. If the second leg falls too — Windows takes a once-in-a-lifetime hit. And the prediction: Microsoft will just try to buy Steam (“if you won’t pay me here, you’ll pay me there”).
• And an ending as it should end — “how many programmers does it take to change a lightbulb” jokes: Unix — three (two write software to download and screw it, the third stands on the switch so no one turns it on), Mac — ten (one writes the software, nine stand next to him with the logo), Microsoft — none (wait for someone else to do it, then buy or steal the solution), and programmers in general — none, because it is a hardware problem. Then a blessing for the people of Israel. That is how the conversation ended.

🔥 Roast Corner

You know what is funny? Everyone talks about “AI for lawyers,” but most tools on the market are just ChatGPT with a prompt template (a pre-written set of instructions) and a new logo. That is it. That is the whole magic.

The real thing is not offering a lawyer “help writing an email.” The real thing is putting 30 documents on the table, reading them, understanding how they connect, calculating odds, finding relevant case law — and packaging it all into a draft you can actually work with.

This is not a chatbot. This is a legal assistant (an actual assistant to a lawyer) that understands context.

And yes, there is a lot of noise: half the startups tweeting “AI for Lawyers” are, in practice, selling you a new window to ChatGPT. But when it is done right — like the demo in the clip — it makes a real difference.

🤖 AI for Humans

For those not inside the legal world — first, a little background: lawyers spend long hours reading documents, correspondence, and court rulings. This is exactly where AI can help:

• Document summarization — quickly read dozens of pages and return the key points.
• Connection discovery — find documents that “talk to each other” inside a large case archive.
• Risk analysis — not an automatic lawyer: a tool that organizes the information and presents options, while the decision stays with the professional.
• Drafts — a first text that can be refined and approved, instead of starting from a blank page.

The tool demoed in the conversation does all of this under one roof — connected to real documents and a real workflow, not another ChatGPT tab.

And here, really, is the big difference: there is AI that knows how to talk — and there is AI that knows how to work.

Published 2026-10-05 from the YouTube conversation between Ariel and Igal.