Mohammed Wasif

tech consulting | building supervisor

[Mohammed Wasif @Mohammed__Wasif]

I got tired of prompting and supervising Claude Code.

So I built Supervisor, a Mac app that reads your corpus, Claude code session repos, generates tailored principles.md’s for each session, and types the prompts into each session itself.

You run 4 terminal and 4 desktop Claude code sessions, and you touch the keyboard zero times.

@addyosmani just wrote about loop engineering. @steipete said stop prompting agents, design loops instead. @bcherny said he has loops prompting Claude for him.

This is what that looks like as software.
Still building but it works.

Comment “Supervisor” 👇 to join the waitlist!

#ClaudeCode #BuildInPublic #LoopEngineering #AI

[Mohammed Wasif @Mohammed__Wasif]

OpenAI’s GPT-5.6 Terra tier matches last quarter’s GPT-5.5 at half the price.

The headlines chase the smartest model, but intelligence is getting cheaper faster than it’s getting smarter 🚀🙌🏽

[Mohammed Wasif @Mohammed__Wasif]

The answer’s a supervisor agent.

[Mohammed Wasif @Mohammed__Wasif]

OpenAI is putting GPT-5.6 Sol on Cerebras at 750 tokens a second.

Holy crap lol

So here’s the question nobody’s answering: when your agent writes code faster than you can read it, how are you actually reviewing what it ships before it merges?

[Mohammed Wasif @Mohammed__Wasif]

When OpenAI launched GPT-5.6, it openly rated it High risk on cybersecurity and bio, then released it behind its strongest safety stack yet.

The frontier race is quietly turning into a race to build enough guardrails to justify shipping what you already built.

[Mohammed Wasif @Mohammed__Wasif]

fable 5 is back!

anthropic's most powerful model got pulled by a us government order on june 12.

amazon researchers found a way past its safety limits and got it to write code for exploiting a software vulnerability. with no way to verify users in real time, access was cut for everyone overnight.

anthropic's own testing showed older models like opus 4.8 and gpt-5.5 could already do the exact same thing. it was never unique to fable 5. they still added a new filter that blocks it over 99% of the time.

[Mohammed Wasif @Mohammed__Wasif]

Fable 5 got pulled by a government letter two weeks ago. In 24 hours it’s back!

But let’s not forget, the most capable model on earth went dark and switched back on, and none of it was your call to make.

This is exactly why the durable part isn’t the model, it’s the loop you run local and own.

[Anthropic @AnthropicAI]

Claude Fable 5 will be available again globally tomorrow.

After a series of productive conversations with the US government, we're redeploying the model with a new set of classifiers to target and block more cybersecurity tasks. In the near term, some routine tasks like coding and debugging will fall back to Opus 4.8. We’ll continue to refine these classifiers over the coming weeks to reduce false positives and better distinguish genuine misuse from legitimate requests.

We’ve also begun drafting a consensus framework—with Amazon, Microsoft, Google, and other Glasswing partners—for assessing the severity of AI jailbreaks and how AI developers should respond to them. We invite other industry partners and model providers to join us in this effort.

Finally, we’re scaling up our collaboration with the US government on model testing and safeguards. This will include pre-release access to models and safeguards for evaluation, information sharing on jailbreaks and misuse, and dedicated resources for joint research.

Thank you to our users for your patience, and to our partners across the government, industry, and the research community who worked alongside us to make Fable 5 available again.

Read our full blog: https://t.co/VHyum831ri

[Anthropic @AnthropicAI]

Claude Fable 5 will be available again globally tomorrow.

After a series of productive conversations with the US government, we're redeploying the model with a new set of classifiers to target and block more cybersecurity tasks. In the near term, some routine tasks like coding and debugging will fall back to Opus 4.8. We’ll continue to refine these classifiers over the coming weeks to reduce false positives and better distinguish genuine misuse from legitimate requests.

We’ve also begun drafting a consensus framework—with Amazon, Microsoft, Google, and other Glasswing partners—for assessing the severity of AI jailbreaks and how AI developers should respond to them. We invite other industry partners and model providers to join us in this effort.

Finally, we’re scaling up our collaboration with the US government on model testing and safeguards. This will include pre-release access to models and safeguards for evaluation, information sharing on jailbreaks and misuse, and dedicated resources for joint research.

Thank you to our users for your patience, and to our partners across the government, industry, and the research community who worked alongside us to make Fable 5 available again.

Read our full blog: https://t.co/VHyum831ri

[Mohammed Wasif @Mohammed__Wasif]

Watch me give Claude a single prompt and supervisor turn it into a finished website! Would you use this? #buildinpublic #claudecode #vibecode #ai

[Mohammed Wasif @Mohammed__Wasif]

Snowflake and Anthropic partnered on what enterprises now call “governed AI.”

It’s smart because at that scale a model with nothing deciding what it can touch is a liability. But is governance enterprise-only?

Anyone leaving an agent running needs a layer watching what it does.

[Mohammed Wasif @Mohammed__Wasif]

Free, open source, Mac only: https://t.co/xJ9tBMlRc6

[Mohammed Wasif @Mohammed__Wasif]

Anthropic has scheduled agents on Claude code. When the timer fires, Claude opens a fresh session and runs on its own, with nobody watching the moment it does what it can’t undo.

Supervisor takes that seat, local on your key.

[Mohammed Wasif @Mohammed__Wasif]

Anthropic shipped Claude Finance which includes 10 pre-built agents that run the analysis for you.

We used this at my client in the account department and realized that you can pre-build the agent but you can’t pre-build the judgment of what’s safe to run unattended on your books, because that part is extremely specific to the entity.

[Mohammed Wasif @Mohammed__Wasif]

The AI model the US government banned is coming back. Quick recap: on June 12 the government pulled Fable 5, Anthropic's most powerful model, over a possible jailbreak. It went dark worldwide overnight. 15 days later it's on track to return. For context, Stripe used it to rebuild a 50-million-line codebase in ONE day. But it's a thaw not a green light, the Pentagon and NSA still have to sign off. Frontier AI is officially a policy variable now. #ai #claude #anthropic #fable5 #aitools

[Mohammed Wasif @Mohammed__Wasif]

Reddit either loved Supervisor or called it AI slop. I read every word, and 30+ comments later V2 is already being built from what you said. Coming very soon, still free, still on Mac. What's the one thing you'd change? #buildinpublic #vibecode #claudecode #aiagents #aitools

[Mohammed Wasif @Mohammed__Wasif]

When @AnthropicAI shipped Claude Security, did anyone else read it as an admission? When agents write most of the code, the risk isn't an outside attacker, it's your own agent shipping something nobody reviewed.

[Mohammed Wasif @Mohammed__Wasif]

@AnthropicAI said it took weeks to catch 28 million automated exchanges quietly draining its own models. If the team that built the model needs that long to spot unwanted automation, your overnight loop needs something watching each action as it runs, not a log you read after.

[Mohammed Wasif @Mohammed__Wasif]

@AnthropicAI's Claude now runs on-device on your Mac through @Apple's Foundation Models framework. The model living on your machine is a real privacy win!

[Mohammed Wasif @Mohammed__Wasif]

The rise and fall of OpenClaw, and the one thing it proved about AI coding agents. Autonomous agents do busywork. Claude Code does real work but needs you babysitting it. Supervisor is the bridge, it runs Claude Code on its own. Now live, free, on Mac. #buildinpublic #vibecode #claudecode #aiagents #aitools

[Mohammed Wasif @Mohammed__Wasif]

Download here!: https://t.co/xJ9tBMlRc6

[Mohammed Wasif @Mohammed__Wasif]

Two of the most capable US models got government-gated this week.

You can’t control the model, but you can own the layer around it.

Supervisor runs that layer locally on your Mac with your chosen API key.