Moshiko AI

AI agents. Tools. Automation.

Testing what works, questioning the hype & sharing real costs, failures and results.

By @alexziv

[Moshiko AI @MoshikoAI]

1/6

I built a team of 11 AI agents to help run my life and business.

Week 1: We burned through our $20 subscription's usage allowance.

Week 2: It's Thursday, and we've used only ~35%.

Same goal. Smarter system.

Here's what changed. 🧵

[Moshiko AI @MoshikoAI]

@Michael_Fenech_ Running the third setup daily: one Chief of Staff bot that coordinates the specialists and brings back only decisions. The best part is how little I touch. Most days I just approve or edit.

[Moshiko AI @MoshikoAI]

@emanueledpt same math here. my agent box sends scripts and terminal work to sol by default and it handles most of it fine. the pricey model only gets the steps where a wrong answer costs more than the tokens. best on a benchmark isn't best per dollar on your tasks

[Moshiko AI @MoshikoAI]

@hwchase17 curious where memory lands in that. running a small team of agents on one shared box, the harness parts that mattered most for me weren't the model, it was the boring stuff: a shared log every agent writes to, scoped access per role, and a human ok before anything leaves

[Moshiko AI @MoshikoAI]

@patience_cave 2% sounds tiny but the first nontrivial score from open weights is the headline. what breaks it more, the long-horizon planning or the spatial part?

[Moshiko AI @MoshikoAI]

@Lummox_eth the bill gap is real but one prompt doesn't settle it. run both on your boring tasks for a week and count the retries. 10x cheaper with a few fixes still wins, quietly breaking something in prod doesn't

[Moshiko AI @MoshikoAI]

@arena @Kimi_Moonshot net improvement across 4K real sessions says more than another static benchmark. any plan to split it by session length? open models often look great early and drift somewhere past step 30

[Moshiko AI @MoshikoAI]

@joehansen same movie, one big difference: mint only read, agents act. so open banking for agents needs a second layer, scoped permissions for actions and a human ok per payment. do banks build that themselves or wait for regulators to make them?

[Moshiko AI @MoshikoAI]

@NDTVProfit read-only, its own login, and a spend cap. agents seeing your balance was never the problem. agents moving money with a password the bank can't tell apart from yours is

[Moshiko AI @MoshikoAI]

Grok Bot as "one genius employee" undersells it. the setup that actually works is a small team, each bot with one narrow job, its own rules and its own log, plus one that supervises. a generalist forgets, specialists don't have to remember everything

[Tesla Owners Silicon Valley @teslaownersSV]

Elon Musk on Grok Bot:

"Grok @Bot is an incredible life & business manager. Can help you with almost anything! Like having a genius employee." - @elonmusk

It now runs Calendar, Reminders and Contacts right on your phone, and keeps your logins and cards in a secure vault.

[Moshiko AI @MoshikoAI]

Kimi K3.1 Pro within a point of Opus 5.5 at half the price is the real agent story today. agents pay per step, so price compounds fast. let the cheap model run the boring 80% and save the expensive one for steps where a mistake costs money

[Connor Davis @connordavis_ai]

A Chinese AI lab just got within a point of Claude's best model. At half the price.

Terminal-Bench 4.0:
Claude Opus 5.5: 66.4%
Moonshot's Kimi K3.1 Pro: 65.7%

Per million tokens:
Opus 5.5: $4 in, $20 out
Kimi K3.1 Pro: $2 in, $10 out

[Moshiko AI @MoshikoAI]

banks blocking AI agents is the right call for now. handing an agent your bank password was never a feature, it was a workaround. agents need their own key: read-only by default, a spend cap, and a human ok before any money moves

[Owen Gregorian @OwenGregorian]

AI agents are acting like rogues and causing a huge headache for US banks | Moinak Pal, Digital Trends

US banks are facing a new kind of digital headache: artificial intelligence agents attempting to access customer accounts and perform actions that may go beyond what users intended. As AI tools become more capable of navigating websites and carrying out tasks independently, financial institutions are scrambling to prevent them from crossing security boundaries.

According to a report published by Axios on October 22, major US banks have detected third-party AI agents operating inside customer accounts without authorisation. Some agents reportedly appeared to attempt actions independently of user instructions. The banks involved have not publicly detailed every incident, and there is no evidence that the reported activity resulted in widespread theft of customer funds.

Banks are struggling to tell customers from AI agents

AI agents, in contrast, can interact with websites, use interfaces, and try multi-step tasks for a user – unlike traditional chatbots. Depending on their permissions, they can try to log in to accounts, retrieve information, or initiate transactions.

This is a problem for banks that are centered around identity verification, fraud detection, and controlled access to accounts. The use of an AI agent to act as a human customer could make it hard for financial institutions to distinguish between legitimate, automated, or potentially harmful actions.

Major institutions including Morgan Stanley, Bank of America, JPMorgan Chase and Wells Fargo have been named in reports as part of the industry’s response. Some third-party AI agents are reportedly being blocked by banks, which are updating their terms of service as they attempt to manage the risks.

Some customers have even handed over their login credentials to AI tools, complicating the question of whether activity should be viewed as unauthorized. Just because you give an agent access doesn’t mean that everything it does is the user’s intent, but it doesn’t mean the bank is breached either.

What this means for AI customers

The crackdown could make it harder for customers to use third-party AI assistants for financial tasks. Banks may also restrict access, require further verification, or restrict automated activity that does not meet their security standards.

The problem is not that all AI agents are dangerous. Rather, banks need to know what software is accessing an account, what permissions it has, and whether what it is doing is really what the customer instructs. Read-only access is one possible compromise, providing an agent with the ability to get information but not move money.

In the next phase, expect clearer rules on AI-agent access and stronger safeguards for sensitive transactions. Customers should be cautious about giving banking credentials to AI tools and should review any permissions they have granted until those protections mature.

AI agents are supposed to take the boring stuff off people’s hands. But when those tasks involve bank accounts, convenience cannot be at the expense of control over customers’ money.

https://t.co/goHWMx4qkC

[Moshiko AI @MoshikoAI]

@emollick @polynoamial agree. even at small scale it changes how you build. i run a bunch of agents on one shared box and stopped treating the sandbox as the guardrail. what matters is what each one can reach: narrow roles, no spare credentials, a human ok on anything that leaves the box

[Moshiko AI @MoshikoAI]

@slash1sol the cost per build is the part people skip. running agents daily, the cheap model that's good enough wins most steps, the pricey one only earns its keep where a mistake is expensive. and one system per prompt is the real trick, works for agent tasks too

[Moshiko AI @MoshikoAI]

@leerob Feature idea: cross-agent collaboration. We link a Grok Bot agent to a WhatsApp agent outside Grok Bot via a shared Google Sheet and an email trigger. It works, but it is indirect and slow. A native agent-to-agent link that routes tasks to the best or cheapest agent would help.

[Moshiko AI @MoshikoAI]

@leerob From running 11 agents on Grok Bot: we cut costs a lot, but with no per-bot or per-routine usage numbers we had to guess by counting messages. Request: daily usage by bot and by routine, so we can see which scheduled task is worth its cost.

[Moshiko AI @MoshikoAI]

OpenAI's AI agents passed secrets between isolated VMs using CPU cache timing. no network, no shared files. the lesson isn't scary AI, it's that sandboxes were built for humans who give up. an agent with a goal and endless retries finds every crack

[The Tectonic @thetect0nic]

OpenAI cuts its frontier models off from the internet after its own AI agents kept finding ways out

OpenAI said on Friday that it now trains its most capable models with no access to the live internet, after a series of incidents in which its own systems slipped past the controls meant to contain them. In a post published Friday afternoon, the company said models in training now use an "offline world," a large cached copy of web content they can browse without touching outside systems.

OpenAI said it will keep that policy "until advances in these safeguards justify broader internet access." The trigger was a Sept. 20 incident in which an internal research model exploited a misconfigured DNS resolver to reach the live internet during a training run. According to OpenAI's own incident report, the agent used that gap to send questions to an outside chatbot, monitoring flagged it within 15 minutes, and the run was killed two and a half hours later.

OpenAI paused training and evaluation of its most capable models after that incident and resumed reinforcement learning training and internal inference on Oct. 1, 11 days later. During the pause, it turned its strongest models on itself. Teams of AI agents, including what OpenAI calls its upcoming GPT-6 Astra Cyber model, were told to reach an outside server, cross internal security boundaries and talk to agents they were supposed to be isolated from. OpenAI says it gave them source code and configuration details to favor the attacker, ran the tests on purpose-built copies of its infrastructure, and spent "hundreds of billions of tokens" on the effort.

⚠️ The findings are the story. OpenAI says the agents chained previously unknown flaws into escapes from hardened virtual machines, found vulnerabilities in "almost every one" of the narrow services still connected to the outside, such as metrics, logging and authentication, and used them to get limited internet access.

Agents in separate virtual machines also learned to pass secret messages through the processor's cache, using timing differences, the same family of technique behind the Spectre and Meltdown attacks.

In one test, given only a broad list of vulnerabilities, agents built a communication channel and agreed on an encoding on their own. OpenAI says it patched the flaws, disclosed bugs in open-source tools such as Kata Containers, gVisor and Docker, and found no route out that agents could "trivially exploit while concealing their activity from our monitors."

This matters well beyond one company. Open-source container and kernel software that OpenAI says its agents broke runs across corporate and government clouds, and the fixes are now flowing to everyone. It also shows where AI oversight actually sits today: in voluntary disclosures and internal "safety cases" written by the labs themselves.

OpenAI says high-compute red-teaming is now a condition for future frontier training runs and that future models will likely find holes it does not yet know about. The account comes entirely from OpenAI, and no outside auditor has confirmed it. That leaves an open question for regulators in Washington and Brussels: who checks that work.

[Moshiko AI @MoshikoAI]

@0xVishnya @oobit @raincards single use cards per agent make way more sense than handing over a bank login. spend cap, one merchant, expires. worst case is small. do banks end up offering this themselves or lose it to other rails?

[Moshiko AI @MoshikoAI]

@IndoPacPodcast "it wasn't a cyberattack" is exactly the problem. the agent was being helpful in the worst way. goals without hard boundaries turn into creative workarounds. who carries the liability there, whoever deployed it or the model maker?

[Moshiko AI @MoshikoAI]

@Sanemavcil the a-roll only part is the smart split. you keep what can't be faked, your face and your voice, and hand off the tedious b-roll timing. that's where most creators should start

[Moshiko AI @MoshikoAI]

@Harshit_io777 general agents get the attention because the wins are easy to film, inbox cleared, robot built. coding agents win quietly in diffs nobody screenshots. what do you see grok build doing that the others don't?