Sonat Yalcinkaya | SoyakaAI

Founder & CEO, SoyakaAI. Building AI for Saudi finance. Sharing what I build and learn with technical builders and finance leaders.

[Sonat Yalcinkaya | SoyakaAI @sonatykaya]

Opus. Sonnet. Astra. Another launch, another decision. Keeping up shouldn’t be your job. I built an engine that picks the model for each job and balances my limits—so one isn’t exhausted while another sits unused. Watch how it routes the work.

I’ve made the starter public so you can use it too. Download it, set your own rules, connect your workers and make it yours.

https://t.co/GAHlgErazI

[Sonat Yalcinkaya | SoyakaAI @sonatykaya]

@derekmross i was just about to get pixel 8 for this swapping to 7pro asap. cannot wait!!!

[Sonat Yalcinkaya | SoyakaAI @sonatykaya]

Holy shit. @Tesla just took the Roadster off the ground. 🤯

Rocket thrusters. An upside-down ramp. A car doing a kickflip. 0-60 in 0.81 seconds

Watch @CNET’s highlights.
https://t.co/e3orCITrGv

#Tesla #TeslaRoadster

[Sonat Yalcinkaya | SoyakaAI @sonatykaya]

You log off. The work keeps moving.

OpenAI Dots puts always-on agents in focus. I already organize work across agents; the opportunity is connecting the handoffs.

Delegate the work. Keep the decisions.

My take + CNET’s demo ↓
https://t.co/lOqZ0IWfJS

#AI #AIAgents

[Sonat Yalcinkaya | SoyakaAI @sonatykaya]

I stopped using Kindle and Audible. Built my own—in one night.

Reverse Engineer Anything is a game changer. Building gets faster; understanding the work matters more.

My take, featuring @natebjones ↓
https://t.co/hjy26v2WK4

#AI #FutureOfWork

[Sonat Yalcinkaya | SoyakaAI @sonatykaya]

Most AI jobs don't need the biggest model. Fixed list → small model. Reasoning → large model. Exact rule → code. In credit, a model scoring 93% on rules is a failure. Build with OpenAI Partner, ed. 2: https://t.co/KYRGCwZCjO

[Sonat Yalcinkaya | SoyakaAI @sonatykaya]

We're now an OpenAI Select Partner in Saudi Arabia. I'm sharing what that work teaches us in a new series, Build with OpenAI.
Edition 1: the three doors into OpenAI (ChatGPT, Codex and the API) and why each needs its own owner.
https://t.co/SfbVDAryYO

[Sonat Yalcinkaya | SoyakaAI @sonatykaya]

We're now an OpenAI Select Partner in Saudi Arabia. I'm sharing what that work teaches us in a new series, Build with OpenAI.
Edition 1: the three doors into OpenAI (ChatGPT, Codex and the API) and why each needs its own owner.
https://t.co/SfbVDAr19g

[Sonat Yalcinkaya | SoyakaAI @sonatykaya]

Appreciate this, @Kantorcodes. You’ve captured it well. Jarvis started with a practical need: one voice entry point for real business workflows, without tying everything to one model. The public starter is there so others can adapt that idea to their own setup. Thanks for pointing people towards it.

[Sonat Yalcinkaya | SoyakaAI @sonatykaya]

I built Jarvis: my own voice command centre.

Not tied to Siri, Codex or Claude. My memory. My rules.

See the Saudi payments → Mudad → Zoho → ZATCA workflow demo.

Get the GitHub starter. Build yours:
https://t.co/W7M7NZEofk

[Sonat Yalcinkaya | SoyakaAI @sonatykaya]

Thanks for the follow-up, @Kantorcodes — and for making the contribution route clear.

That’s the idea behind Jarvis: a self-hosted voice layer with local speech and agent routing, while keeping memory and workflows under your own control. The models can change without having to rebuild everything around them.

A small starter people can adapt to their own work matters more to me than another assistant they’re locked into. Development & Workflow sounds like a good fit.

[Sonat Yalcinkaya | SoyakaAI @sonatykaya]

Hands down, Opus 5.5 for builds. Some of what it can do is insane. 😄

And thank you, @PovilasKorop — your videos are a big part of how I keep up with these models. You do an amazing job making the differences visible.

Keeping up is almost a job in itself now. That’s why we put together a model selector: match the task to the model and effort level, check how much quota is left, and adjust the routing. That kind of coordination becomes really useful when several agents are working together. https://t.co/20nHEB2EUP

But there’s another shift alongside the frontier models that has been huge for me: Jev and System 1 decision-making. Learning how a business makes its recurring judgments, then helping with those decisions, changes the day-to-day work in a very different way. Especially in B2B.

So yes, Opus is my pick for building. The bigger picture is knowing which kind of intelligence the job actually needs.

[Sonat Yalcinkaya | SoyakaAI @sonatykaya]

Opus. Sonnet. Astra. Another launch, another decision. Keeping up shouldn’t be your job. I built an engine that picks the model for each job and balances my limits—so one isn’t exhausted while another sits unused. Watch how it routes the work.

I’ve made the starter public so you can use it too. Download it, set your own rules, connect your workers and make it yours.

https://t.co/GAHlgErazI

[Sonat Yalcinkaya | SoyakaAI @sonatykaya]

This matches what I’m seeing in my own builds. Opus 5.5 is still my first choice.

But the best model depends on who you are and what you need it to do. A developer’s coding test doesn’t represent every Google or Meta user, or every social, WhatsApp and everyday workflow.

I built a model selector for exactly this reason: Opus for my builds, Astra/Codex and Dots for other parts of the workflow, and Jev for repeatable System 1 judgments in B2B transformation.

Different jobs, different tools. Opus winning my building workflow doesn’t make the others useless. Know your use case, then choose your model.

[Povilas Korop | AI Coding & Laravel Educator @PovilasKorop]

My prediction what LLM developers will actually use after this week's launches so far.

[Sonat Yalcinkaya | SoyakaAI @sonatykaya]

✨ BREAKING: Seven major Adobe-style creative apps—rebuilt as free, open-source alternatives.

Photoshop. Illustrator. Premiere Pro. Lightroom. Acrobat. After Effects. InDesign.

ArtCraft covers all seven. You can download the tools, access their source code and customise them around your own workflow. AI agents can work inside them too.

For me, this signals a seismic shift: more people and organisations can shape the software they need to unleash their own capabilities.

▶️ See the walkthrough from FUTC:
https://t.co/H63iotFGUP

📌 Next, I’ll share how we applied the same thinking at SoyakaAI—replacing our company knowledge hub with our own reading, listening and learning experience, bringing together features we associate with Kindle, Audible and Udemy. I’ll show how you can build something similar for your team.

✨ باختصار: سبعة بدائل مفتوحة المصدر لأبرز برامج Adobe، يمكن تخصيصها حول احتياجاتك والعمل داخلها بمساعدة الذكاء الاصطناعي. التحول هنا هو قدرتنا على تشكيل أدواتنا وإطلاق إمكانات فرقنا. قريباً سأشارك كيف طبّقنا الفكرة لبناء مركز معرفة وتعلّم خاص بشركتنا.

#AI #OpenSource #FutureOfWork

[Sonat Yalcinkaya | SoyakaAI @sonatykaya]

BREAKING: AI can now help reverse-engineer software and recreate features from existing apps and games.

In plain English: parts of an expensive product you once had to buy can now become something you build yourself—with AI.

If you run a company or invest in SaaS, pay attention.

My thesis: when functionality becomes easier to reproduce, features alone become a weaker defence for premium subscriptions and valuations.

The question is shifting from “What does this software do?” to “What does this company have that’s actually difficult to replace?”

That’s why I follow what builders are doing—not just what gets presented in boardrooms.

Watch this breakdown from AI Search:
https://t.co/YfRHjoMICN

[Sonat Yalcinkaya | SoyakaAI @sonatykaya]

This week I moved my AI work to an always-on server.

Herdr brings my chats together. Shahi lets me check agents from my phone. Background jobs keep running when I close the laptop.

Close the laptop. Keep the work moving.

Workflow illustration.

[Sonat Yalcinkaya | SoyakaAI @sonatykaya]

@lukesophinos @threadreaderapp unroll