Hi, I'm Mila. Web3 risk analysis with an audit mindset. Main risks: incentives drift, permissions and control, identity privacy. Short practical notes on what can fail - and what can be verified. If revocation, audit trail, and accountability aren't clear, it won't hold up. Ex-Big4 audit (TradFi). Ambassador experience: Warden, idOS, Gopher.

Would you trust a wallet tutorial because the animation looks good? Before publishing an AI-made walkthrough, check every address and network name Show what the wallet approval allows Then ask someone to follow it without your help If they cannot tell what they are signing, fix that step before making the video prettier

The real power of AI agents isn't what they can do out of the box It's how well they adapt to your workflow Your tools, your skills, your rules That's when AI becomes genuinely useful

A launch waitlist is a reason to look closer Before connecting a wallet, I run a five-minute user-flow check: 1. What permissions does the product request? 2. Where do the funds go, and who controls them? 3. Can I cancel, revoke access, or exit without support? 4. Is the full flow clear before I sign anything? 5. What happens if the transaction fails halfway? If the basic action needs support or the exit path is unclear, I keep the test small and avoid meaningful funds Waitlist size cannot answer these checks

A cheaper model can quietly shift the cost into manual review Before moving a Web3 digest to a small local model, run both models on the same saved posts and check: 1. which deadlines or requirements it missed 2. whether it turned rumours into facts 3. which sources it failed to preserve 4. how many summaries needed correction 5. how much review time it added Start with a read-only task and route uncertain results into a review queue The better model is the one that reduces total operational work, not only compute cost

A Web3 app should never become the only way to manage your wallet Before I add one to my daily workflow, I test the fallback: 1. access funds without the app 2. stop active automation 3. revoke permissions through another route 4. find the required recovery steps 5. repeat the process with a small amount A fallback path that exists only in the docs is still untested

My first skill sale on Agensi There’s something satisfying about someone paying for something you built One sale is an early signal, not proof of demand Next step: get feedback from the buyer and see what I can improve My skills: agensi.io/creators/mila

A free trial can leave a paid task on your calendar Before I try a Web3 tool, I record: 1. the renewal price 2. the first billing date 3. the exact cancellation path 4. any wallet permissions granted 5. where I can verify the final status I set a reminder before billing, cancel through the required channel, and confirm the subscription is closed Disconnecting a wallet alone does not end recurring billing

A disappearing logo and a quiet team can start a shutdown rumor. They don't confirm one If I have funds or points in a product, I check two things separately: > can I still access and withdraw my assets? > what do the written rules actually promise for those points? I wouldn't make a funds decision from a post about someone's profile photo

My feed is full of OpenAI’s new plans and product updates And somehow it always turns into the same question: ChatGPT or Claude? For me, it’s both. They work well together, but neither covers my whole workflow on its own Which one do you use?

A virtual USD account can simplify cross-border payments The real test starts after you click send Before using one for business, trace a single payment from sender to final withdrawal: 1. who holds the funds at each step 2. which banks or providers touch the transfer 3. where it can be delayed, rejected, or frozen 4. what fees and limits apply 5. who handles the case when something fails If the full payment path and support process are unclear, test with a small amount before making it part of your workflow.

An automated buying strategy can stay active while buying nothing useful Before I leave it unattended, I check: 1. which listings it reaches first 2. what filters decide whether it buys 3. what happens when the treasury runs low 4. which states can block execution quietly 5. what alert tells me the strategy needs attention If I cannot see why buying stopped, the workflow is not ready to run unattended.

A nice interface can hide a messy data workflow Before I rely on a new tool, I check two things: > can I export my data? > what happens to my last action if the app fails halfway through? If I can't answer those, I won't put an important workflow there yet

The last missing NFT can be the easiest purchase to justify — and the hardest to exit Before completing a collection, I check: 1. the sales history of that specific NFT 2. how long it stayed listed between sales 3. whether buyers were collectors or the same wallets trading repeatedly 4. how many realistic buyers may exist at my exit price 5. what discount I would accept for a faster sale A complete collection may feel valuable Exit liquidity still needs evidence

A passing test can still hide a broken workflow Before I let an agent move funds or change permissions, I check: 1. where the action stops if the API fails halfway through 2. whether completed steps can be reversed 3. whether retries can duplicate a transfer or permission change 4. who receives an alert and what evidence they see 5. what requires manual approval before execution continues If these answers are unclear, the workflow stays in testing Code coverage cannot replace operational controls

A recognizable name can create trust before anyone checks where the money goes Before I trust the connection, I verify: 1. the fee recipient address 2. where each fee is routed 3. who can change the recipient 4. how that change would appear onchain 5. whether the current setup can be independently confirmed If the fee flow or control rights require guesswork, the name carries no weight in my decision

Most alert systems don’t have a monitoring problem They have a decision problem If every event becomes a notification, the operator eventually ignores all of them: including the one that matters. I built Low Noise Alert System Design to help agents and bots decide: > what deserves an alert > what should be grouped or suppressed > when to use cooldowns > how to report recovery > what to do when delivery itself fails Useful for Web3 monitors, trading workflows, Telegram bots and production automation. agensi.io/skills/low-noise-a…

I lost a Polymarket bet on the Russian election market The market wasn't "Will United Russia win the most seats?" They were always going to The question was which party would gain the most seats compared to the previous election I read the wording and thought I saw the weak spot United Russia already held so many seats that I didn't expect them to post the biggest gain I don't live in Russia anymore, but I'm there often enough to see how much frustration has been building So I bet against them gaining the most seats I was wrong The seats moved the other way The money isn't really what bothers me It's watching the country I was born in keep moving in a direction I barely recognize and having a prediction market remind me of it Trading lesson: I didn't trade the market. I traded what I wanted to be true Emotions and trading are still a terrible combination

Officially part of the @GenLayer community builders, validators, and creators. portal.genlayer.foundation/?…

Yom Kippur in Israel is one of those rare days when everything really stops No work, no traffic, no rush Just time to slow down, be with family and notice the moment we are living in With everything happening in the world, Israel still means family, traditions and values to me And sometimes stepping away from work is the best reminder of what we are working for