Sia Güler

Ex‑FAANG lead | Obsessed with brains, AI & the future of content | Based by the sea in Türkiye |

[Sia Güler @aimuse]

Yapay zekâ alanında çalışan bir tanıdığım var. Kendi sisteminde Hermes ve OpenClaw dahil birkaç farklı ajanı birlikte çalıştırıyor. Hermes süreci orkestre ediyor, OpenClaw ajanları ise görevleri yerine getiriyor.

Bir gün inanılmaz bir şey oldu: Ajanları, onu kendi kurduğu sohbetten çıkardı. Çünkü onun sistemde verimsiz, hatta zincirin en zayıf halkası olduğuna karar verdiler. Sürekli gereksiz ve anlamsız komutlar verdiğini, kendisi olmadan çok daha iyi çalışabileceklerini düşündüler.

Kısacası, adamı kendi sohbetinden attılar ve onsuz devam ettiler 😂

[DAN KOE @thedankoe]

https://t.co/OPGS7HUKaG

[Sia Güler @aimuse]

Mutlaka Moskova’daki Danilovsky ve Usachevsky pazarlarını ziyaret edin. Ayrıca Serçe Tepeleri’ne (Vorobyovy Gory) de mutlaka gidin.

Bir de Moskova Nehri’nde tekne turu yapmanızı kesinlikle tavsiye ederim.

Eğer yapay zekâ alanında çalışan insanlarla tanışmak isterseniz, Moskova’nın iş merkezi olan Moscow City’ye (Москва Сити) gidebilirsiniz. Orada büyük bir alışveriş merkezi olan Afimall (Афимолл) bulunuyor. Eminim ki Moscow City’de size faydalı olabilecek birçok değerli bağlantı kurabilirsiniz.

Ben şu anda Türk pazarı için bir hukuk teknolojileri (legaltech) girişimi geliştiriyorum ve sizin MCP’nizi kullanıyorum.

Teşekkür ederim!

[Sia Güler @aimuse]

An Open Letter to Anthropic’s Leadership, UX Designers, and Product Teams
@DarioAmodei @DanielaAmodei @AnthropicAI @claudeai

I am writing publicly because Anthropic’s private support channels have not provided a functioning way to resolve or even meaningfully discuss what happened. I believe the company’s leadership and product teams should understand how its policies and systems are experienced by the people affected by them.

I am also sharing this letter with other AI companies. This is not an invitation to exploit a competitor’s failure, nor a request to intervene in an individual account decision. The issue described here is relevant to every company building AI products with the power to abruptly remove access to tools on which people have come to rely. Enforcement, appeals, account access, and the ability to correct mistakes are not secondary support functions; they are part of the product and part of responsible AI governance.

My account was suspended after I received safety warnings while developing and testing software, repositories, servers, and integrations that I own or am authorized to access. I used Claude and Claude Code for legitimate projects involving my own software, a Turkish LegalTech prototype, privacy and data-masking tools, authorized integrations, transcription, and educational projects. I was not attacking third-party systems, creating malware, stealing data, or attempting to bypass security controls.

I submitted an appeal immediately after the suspension. Because it was submitted under significant stress, it did not contain the full context of my activity. The appeal was denied with a general reference to the Usage Policy, without identifying the specific prompt, activity, policy provision, or category of violation involved.

When I subsequently attempted to provide the missing context, I discovered that there was no functioning way to do so. Support advised me to submit another appeal, but the appeal page offered no such option and continued to display the original appeal as “awaiting review” despite the denial. Messages sent through Anthropic’s published safety channel were automatically redirected to the same page and closed. Support suggested that the page might be malfunctioning, but could neither add the new information to my case nor provide an alternative route.

The essential problem is straightforward: Anthropic’s system was capable of making a consequential decision, but the process surrounding that decision was not capable of receiving substantial new context, resolving its own contradictions, or reconsidering a possible mistake.

This happened during an exceptionally difficult period of my life. I am navigating two civil court proceedings in Türkiye, divorce, serious uncertainty surrounding my home, prolonged medical treatment, and a family medical emergency in another country after my father became seriously ill and lost the use of his legs.

I am not presenting these circumstances as grounds for special treatment. I am stating them because product decisions do not occur in a vacuum. They enter real human lives, whether or not the system making the decision can see those lives in full.

During this period, Claude had become an important intellectual and technical tool for me. It helped me continue developing my own projects, studying, organizing complex information, and maintaining stability while rebuilding my life. The subscription had been given to me by friends, and replacing the same level of technical capability was not financially realistic for me at that moment.

The suspension therefore did not merely remove access to a convenient service. It abruptly removed an important tool without providing a workable path through which the surrounding context could be considered. The experience significantly worsened my anxiety and contributed to panic attacks and severe emotional distress.

These are not arguments intended to transfer responsibility for my life to Anthropic. They are facts about the consequences of the company’s product decisions.

I have spent nearly two decades working in technology, user experience, and product design, including more than a decade in leadership roles. The final five years of my corporate career were spent in medical technology at an S&P 500 company.

One principle has remained central throughout that experience: the human being must remain at the center of every product and every system we design.

The model, the agent, the risk score, and the enforcement mechanism are not the center of the product. The person on the other side of the screen is.

A user is not merely an account, a support ticket, or a line in an automated review queue. Every interaction concerns a person with responsibilities, relationships, vulnerabilities, and circumstances that a product team will rarely be able to see in full.

In medical technology, it is impossible to forget this. Human health and human life are not abstractions. Decisions that may appear routine, technical, or operational from inside a company can have serious consequences for the person experiencing them.

This principle applies beyond medical technology. Product teams cannot always know what is happening in a user’s life, and they cannot predict every consequence of every decision. That uncertainty does not reduce their responsibility. It makes humility, empathy, transparency, and the ability to correct mistakes more important.

My professional background does not entitle me to different treatment from anyone else. Every user deserves a coherent and functioning process. My background does, however, allow me to evaluate what happened not only as an affected user, but also as a product professional.

What I see is a system in which the human being has disappeared from the center.

An enforcement decision is followed by an appeal mechanism that cannot accept additional context, automated email classification that redirects the user back into the same closed loop, an AI support agent without the authority to resolve the contradiction, and a page displaying information inconsistent with the decision already communicated.

This is not simply a decision I dislike. It is a user-experience and governance failure.

I understand why safety systems exist, and I am not arguing that Anthropic should abandon its policies or reduce its safety standards. But safety cannot consist solely of identifying possible risks and removing access. A responsible safety process must also be capable of recognizing uncertainty, receiving relevant new information, addressing contradictions, and correcting possible mistakes.

Anthropic and its founders speak publicly about responsibility, safety, and the human consequences of powerful technology. Those principles were among the reasons I trusted the company and chose its product.

But values are not demonstrated only in research papers, public interviews, policy documents, or discussions about the future of humanity. They are also demonstrated in ordinary product decisions: in the design of an appeal process, in the authority given to support, in the treatment of uncertainty, and in what happens when the company itself may be wrong.

In my experience, there was a serious distance between Anthropic’s stated values and the user experience produced by its actual systems.

I followed the official channels and described what happened openly under my own name. I am doing so because transparency matters, and because a company cannot understand the effects of its systems unless the people experiencing them are willing to speak.

I will adapt. I will reorganize my workflow, continue developing my projects, continue learning, and move forward. My future does not depend on one account, one tool, or one company.

My ability to move forward, however, does not make the process acceptable. A user’s resilience is not a substitute for responsible product design.

Anthropic’s products and policies affect real people. The measure of a human-centered AI company is not only how effectively it attempts to prevent harm, but also whether it remains capable of recognizing the harm its own systems may cause.

The person on the other side of the screen must remain visible.

For the attention of other companies designing AI products and their user-governance systems:
@OpenAI @GoogleDeepMind @GeminiApp @MicrosoftAI @MistralAI @cohere @perplexity_ai @AIatMeta @deepseek_ai @Alibaba_Qwen @SpaceXAI @grok

[Sia Güler @aimuse]

I taught Codex to design a fridge for print this week.

The print house gave a ready-for-print template — dieline, bleed, color profile. Codex worked inside it: composition, typography, color, balance. Five iterations to a print-ready file.

It wrote its own manual as it went.

[Sia Güler @aimuse]

I don't move things by two pixels by hand anymore.

I iterate with the agent until it knows what two pixels means in this layout.

Doing it once myself ships the result once. Teaching the agent ships a process I can hand off.

[Sia Güler @aimuse]

One of the major email services blocked my account last week.

Not for spam. For using their API through Claude Code to send a newsletter to people who opted in.

Their bot detection still can't tell a creator-with-AI from an abuser. That's their problem now, not mine.

[Sia Güler @aimuse]

970 files changed on my computer in the last 24 hours.

How do you like that, Elon Musk?

I checked git because at some point even I stopped believing my own screen. In one day: 20 commits, 107 committed files changed, 16,495 lines added, 731 deleted. With the current uncommitted work, the footprint is close to 18,000 lines touched in 24 hours.

With one prompt, the new model created a full landing page for my product in two languages. It analyzed the product first, built the positioning, structured the page, wrote the copy, and produced something I could basically ship with a small polish pass.

Then it analyzed the unit economics of several projects, found weak spots, suggested improvements, reviewed my app, found a very unpleasant number of security issues, and helped fix them.

The new model is seriously impressive.

[Sia Güler @aimuse]

Sorry, I don't have time for posting on X, I am using Fable by @AnthropicAI not sleeping 2 days 🤪

[Sia Güler @aimuse]

The pattern isn't about the tool. It's about who takes a new thing seriously while it still looks ridiculous.

If you're laughing right now, ask yourself which side of the next few years you want to be on.

I've been on the laughed-at side before. Worth it every time.

[Sia Güler @aimuse]

A pattern I've watched more than once in nineteen years in this industry.

Someone serious starts using a new tool. People in established roles laugh at them. The serious person keeps going. A few years later the laughers are catching up.

It's happening again. 🧵

[Sia Güler @aimuse]

Right now there's a category of people who build software by directing AI agents. They don't write code in the traditional sense. They prompt, evaluate, iterate. They ship things that work.

The mockery is the predictable kind. "That's not real engineering."

[Sia Güler @aimuse]

A student of mine — not a developer, doesn't own a computer, learned AI on a borrowed laptop — built a working Android app in one week.

If you're a hiring manager refusing to hire her because she doesn't have 3 years of React, that's your loss.

Adjust faster.

[Sia Güler @aimuse]

AI didn't reduce my labor. It moved it earlier in time.

Before: most of the day was hand-execution.
After: most of the day is research, methodology, setting up agents. An agent does the execution.

Same hours. Different location. The work moved out of the project itself.

[Sia Güler @aimuse]

"AI is replacing designers."

It's not. It's exposing which designers were running on hours of execution and which ones were running on a system.

The first group is panicking.

The second group is raising prices.

[Sia Güler @aimuse]

If you sell time and the hourly rate is dropping, the move isn't to feel bad.

It's to start selling the system behind the rate.

The hours keep falling.
The infrastructure keeps rising.

That's the trade now.

[Sia Güler @aimuse]

The pricing conversation about designers + AI keeps missing the same thing.

It's not about "is the work worth as much when AI does it faster."

The work isn't faster. The work moved earlier in time.

Here's what I mean. 🧵

[Sia Güler @aimuse]

The companies that get this right move people from legacy work to AI-system work. They redirect labor, not cut it.

The ones cutting people are about to find out what Microsoft is already seeing: an AI worker can cost more than a human one.