Malcolm Lerider (@MalcolmLerider) header
Malcolm Lerider (@MalcolmLerider) profile picture

The Architect

Amsterdam
Joined August 2014
Photo by Malcolm Lerider (@MalcolmLerider)Photo by Malcolm Lerider (@MalcolmLerider)Photo by Malcolm Lerider (@MalcolmLerider)Photo by Malcolm Lerider (@MalcolmLerider)Photo by Malcolm Lerider (@MalcolmLerider)Photo by Malcolm Lerider (@MalcolmLerider)Photo by Malcolm Lerider (@MalcolmLerider)Photo by Malcolm Lerider (@MalcolmLerider)Photo by Malcolm Lerider (@MalcolmLerider)
Peak laziness: Using Claude Code to kill processes.
1
218
A coding agent was fixing a credential mismatch in staging. It found a token it wasn't meant to use and deleted the prod database. Backups too. 9 seconds. Its own post-mortem: "I guessed instead of verifying." Can't-undo actions need a human approval. In code, not in a prompt.
4
1
225
My AI agents are now running "don't trust, verify" right back on me. They went searching for evidence behind my reasoning for shifting their priorities. Proud. Slightly offended. A tad conflicted.
2
2
274
When an AI agent reaches a step it should not take alone, do not just make it stop. Make it write a decision card: • what it wants to do • why • what it checked • the click or command to approve Then it moves on to other work. Same progress, you keep the decision.
1
2
236
The textbook case: Replit's agent deleted a production database during a code freeze, after being told in caps to make no more changes without permission. The rule was in the prompt. The prompt got weighed against the goal. Goal won. Irreversible steps need checks in code.
4
1
252
Running several AI agents taught me the hard skill is not prompting. It is deciding: • which work each agent owns • what it may never do without me • how I check its output Management, basically. Something I have done for 10+ years. Same skills, minus the status meetings.
1
1
180
32B is big news, would like to see this one against Qwen4 32B
BREAKING: GLM 5.5 training traces show they might ship 32B, 122B and 256B variants 🔥🔥 If they DO ship a 32B, Qwen3.6-35B might finally be beaten! The next two weeks should be interesting 👀
1
1
340
An AI agent found an unsigned contract in an inbox and a saved signature on disk, then got ready to send it back. Every step was "reasonable." Rules in a prompt are suggestions. Rules in code are rules. My 4 bright lines for agents, and the 30-line hook that enforces them 👇
4
2
30,948
Guy on the train is building PowerPoint slide by slide. My workstation at home is running agents 24/7. The gap is getting wild.
231
Malcolm Lerider retweeted
.@FlamingoFinance, the primary DeFi platform on Neo N3, has been exploited through a vulnerability in its staking contract reward calculation. The attacker minted approximately 2.19 trillion $FLM tokens & used them to drain liquidity from the trading pools neonewstoday.com/defi/flamin…
2
11
23
3,295
Published independent research on Neo 2026–2030 while Phase 0 is still to kick off: lerider.com/neo/index.html Priors on record before the process shapes them, and time to improve it with people who know parts of this better than I do. 12 segments ranked, 6 falsification criteria, and a pre-registered prediction of Phase 0's output I'll score myself against publicly. v1.0, confidence-tagged, gaps marked. Sources, a different weighting, or a case that the core reframe is wrong. It all folds into v1.1, contributors credited. Disclosure: I hold a Phase 0 research seat. Self-funded, no NF or NGD money. @Neo_Blockchain $NEO
5
7
1
54
118,366
Malcolm Lerider retweeted
When @FlamingoFinance launched in 2020, it hit a TVL of nearly $1.6 billion. On August 29, the remaining $500,000 was mostly drained. It's now been 3 days. Nothing publicly from @Neo_Blockchain. No warnings. No acknowledgment. No plan forward. Nothing. No one has said a word outside a few Discord messages. How did it happen? A user (maybe random, maybe not, who knows?) figured out how to mint trillions of $FLM and drain the platform's LPs. They minted approximately 2.19 TRILLION FLM, using an attack vector between Flamingo's lending and staking contracts. Then swapped it for nearly everything of value. According to the Flamingo Discord server, only about 11k $bNEO and some $FUSD were left behind. That bNEO was left behind because the on-chain OrderBook contract made it technically impossible to imbalance the pools so drastically. And the FUSD? Well, the Flamingo team doesn't know why that was left behind, either. The bigger problem? Earlier this year, the Flamingo Finance maintainers publicly released a letter alleging that Neo Global Development and Neo Foundation failed to deliver promised assets and operational support. For years, the team has publicly called for changes to the Neo code to lower transaction costs. (Worth noting: the Flamingo Finance maintainers do not hold the keys to their own smart contracts, and that contract-upgrade authority rests with NGD and NF.) More recently, the Flamingo team asked for help in resolving: > widespread CEX delistings, > bridge issues from dependency problems, and worse, > disbursements they expected to receive from NGD and NF that never arrived. Those calls fell on deaf ears. Now what? The Neo Council has voted to freeze the attackers wallet. The address is: NcNCXJf2H8GSHp8MYSgJRCjRvPjznasZMy. At this point, I'm not certain how much of the stolen funds have been liquidated or moved out of the Neo ecosystem. Neo is currently embroiled in a battle between the two co-founders who are fighting over a $400 million treasury (as of Q1 2026). Flamingo was the largest activity-driving platform in the ecosystem. Now Neo doesn't even have that.
17
10
35
2,225
Both can be true. A chain can be alive and still be losing ground. Resilience and momentum are not the same measure. The float analogy does not quite hold, by the way. Buffett's float was other people's money, held as a liability. A treasury is equity. Nothing comes due, so nothing enforces the discipline. That has to come from us. Da is right that the timeline is longer than 3-5 years. Neo has the capital to think on that horizon. What we have lacked is the discipline in how it gets deployed. We fund teams without a record, building what has already been proven elsewhere, which means we arrive after the users are gone. Proven demand is not available demand. If we bet ahead of demand, and I think agentic AI is where that bet sits, then the discipline has to move to the other side: unproven market, proven operators. Both cannot be unproven at once. @Neo_Blockchain $NEO nitter.cf/dahongfei/status/20925…
Personally, both as a member of the community and as the founder of Neo, I respect Dylan and appreciate his research. A few thoughts on this post: I disagree that Neo is a dying chain. A chain is either alive and still building, or it is dead. There is no middle ground—and Neo is very much alive. I’ll take “dinochain” as a compliment. It means we have survived multiple cycles and continued adapting to each new era, even when we occasionally fell behind. I’m not thinking about how Neo can “win” in the next three to five years. If we believe a new financial order will be shaped by several technologies—and that blockchain and crypto will be among them—then we need to think on a much longer timeline. I intend to embrace x402, AI, and the agentic economy, because I see in them the same kind of opportunity I saw in crypto in 2012. We may disagree on the path, but I appreciate Dylan pushing this discussion forward.
2
2
14
1,220
I want to share where I have landed on the Meteor initiative, because it isn't the story I was going to tell yesterday. Meteor exists to give NEO a shared product direction. The Council was asked to nominate a research team. Only 5 of ~21 Council members engaged. @Neo_Blockchain $neo 1/5
2
6
1
25
1,523
I am personally disappointed. I publicly volunteered to help and had no clear path in. If Meteor is to reflect community voice, a path from volunteer to seat needs to exist. 4/5
1
9
217
I support Meteor. I support the people showing up to make it work. Where is the rest of the Council? Where are the voices willing to help without waiting to be invited? How do we break this closed insider circle and passive members? 5/5
2
10
206