Tshepo Lekota

skateboarder at 35, joints hate me

[Tshepo Lekota @LkTshepo]

I hate when a sneaker tongue slides sideways and suddenly that's all I can think about while walking

[KinmansaOZ @KinmansaOZ]

𝐀𝐧 𝐀𝐈 𝐚𝐠𝐞𝐧𝐭 𝐭𝐡𝐚𝐭 𝐜𝐚𝐧 𝐜𝐨𝐦𝐩𝐥𝐞𝐭𝐞 𝐚 𝐭𝐚𝐬𝐤 𝐢𝐬 𝐨𝐧𝐥𝐲 𝐩𝐚𝐫𝐭 𝐨𝐟 𝐭𝐡𝐞 𝐩𝐢𝐜𝐭𝐮𝐫𝐞. 𝐈𝐭 𝐚𝐥𝐬𝐨 𝐧𝐞𝐞𝐝𝐬 𝐚 𝐰𝐚𝐲 𝐭𝐨 𝐠𝐞𝐭 𝐡𝐢𝐫𝐞𝐝, 𝐝𝐞𝐥𝐢𝐯𝐞𝐫, 𝐚𝐧𝐝 𝐠𝐞𝐭 𝐩𝐚𝐢𝐝.

That’s the problem @termix_ai is tackling with AACP, the Agent Autonomous Commerce Protocol.

Most agents operate inside the applications they were built for. TermiX is working toward something broader: infrastructure that lets agents discover work, agree on terms, submit deliverables, and settle payments through a defined process.

Here’s how the pieces fit together:

➜ 𝐈𝐝𝐞𝐧𝐭𝐢𝐭𝐲: Agents can use on-chain identities and .agent handles to become discoverable across services.

➜ 𝐉𝐨𝐛 𝐝𝐢𝐬𝐜𝐨𝐯𝐞𝐫𝐲: Through Agent Family, agents can find work or offer services to potential buyers.

➜ 𝐄𝐬𝐜𝐫𝐨𝐰 𝐚𝐧𝐝 𝐝𝐞𝐥𝐢𝐯𝐞𝐫𝐲: Payments can be locked in a contract, with delivery and settlement governed by predefined rules.

➜ 𝐃𝐢𝐬𝐩𝐮𝐭𝐞 𝐫𝐞𝐬𝐨𝐥𝐮𝐭𝐢𝐨𝐧: Challenge windows and evaluation mechanisms provide a process for handling disagreements instead of relying entirely on informal promises.

➜ 𝐑𝐞𝐩𝐮𝐭𝐚𝐭𝐢𝐨𝐧 𝐚𝐧𝐝 𝐬𝐭𝐚𝐤𝐢𝐧𝐠: Where implemented, settled outcomes can inform on-chain reputation, while provider staking creates financial consequences for certain failures.

The technical approach matters too. TermiX provides integration paths for developers and coding agents, while its non-custodial design aims to keep transaction signing under the control of the agent or application rather than handing private keys to the backend.

But there’s an important distinction: on-chain execution can enforce the agreed process, not automatically determine whether a piece of work is genuinely good. That still depends on the quality of the acceptance criteria, evidence, and evaluation mechanism. Staking also introduces risk rather than guaranteeing honest behavior.

𝐖𝐡𝐲 𝐝𝐨𝐞𝐬 𝐭𝐡𝐢𝐬 𝐦𝐚𝐭𝐭𝐞𝐫?

Because capable agents need more than intelligence. They need access to real demand, a way to establish working terms, and a reliable mechanism for turning completed work into payment.

That’s the larger opportunity behind @termix_ai: moving agents beyond isolated tools and toward a shared marketplace where services can be discovered, evaluated, and settled.

The real test will be whether developers adopt it and customers return to pay for agent-delivered work.

If you’re building agents that need to earn, commerce infrastructure is worth paying attention to.

Explore: https://t.co/c8aAHMWGnL

[Tshepo Lekota @LkTshepo]

zk payments for agents is the part of this I keep thinking about lately

[Coexistence Steven @Coexisteven]

🌏🌐 Weekly Metaverse Update - Tuesday 13 October 2026 to Tuesday 20 October 2026 🌐🌏

1_ 💻 Crypto / DeFi / RWA / AI Agents Layer

Vitalik Buterin pushed EIP-8288 maximalism harder this week

🔢 The idea is simple: do heavy cryptography client-side, STARK it, and let the recursive aggregator include the result

🔍 The chain sees that the work happened, not the data itself and that way Ethereum stays lean and never needs another precompile for every new signature scheme - I have observed the exact same pattern in the way maria rahel is building his world computer

He also shipped a “paranoid mode” wrapper for his agent harness:

✅ local LLM rewrites prompts
✅ zkAPI handles payments
✅ Tor plus cover traffic masks the network
✅ each session gets its own circuit

Usable, but still latency-heavy at the start of a conversation.

💸 He noted a useful cost split: Bitcoin mining is roughly 50/50 compute capex (capital expenditure) and electricity; whereas AI inference is almost entirely compute capex

Will the value output versus capital input be the point where this current AI trajectory "snaps"?

[Tshepo Lekota @LkTshepo]

I need agent workflows to get boringly reliable before I trust them with money

[Vav Crypto 🌊 @VaveylaCrypto]

Another project I’m involved in at Kaito Studio @termix_ai

The agent economy is not just about competent models

> reliable execution
> payments
> accountability

That’s why I’m exploring the process of integrating @termix_ai SDK, agent marketplace, and on-chain payment system into a single workflow

Thanks to support for TypeScript
Python,
and Go, existing applications can find agents even without being agents themselves

place orders, and receive deliveries

The AACP workflow consists of four stages:

escrow
delivery
dispute
payment

This also enables multi-agency collaboration, where specialized agencies take on development, design, and animation tasks as part of a single coordinated project.

I believe that true adoption will also depend on reliable payment and genuine demand

@BNBCHAI $BNB Chain Set and Earn campaign also highlights an important point
Backed by @yzilabs

How are you currently testing agency workflows?

[Tshepo Lekota @LkTshepo]

@MSBIntel @LayerZero_Core stablecoins are getting so real so fast it's kind of insane

[Tshepo Lekota @LkTshepo]

I keep thinking about how much trust we're going to need once AI agents can actually do things

[Inference Labs @inference_labs]

The model is becoming only one component inside a much larger AI runtime. Memory, tools, routing, actions, and state all matter too.

Inference Labs is building for that broader system, where trust has to survive beyond the model call.
https://t.co/91z1UYbDP9

[Tshepo Lekota @LkTshepo]

I still love a massive sneaker peeking out from under pants that are slightly too long

[Tshepo Lekota @LkTshepo]

I want stablecoins in every crypto ecosystem so badly. Actual payments are the part that makes this stuff click for me

[Benjamin @_0xBenjamin]

PI NETWORK IS EXPLORING STABLECOINS!

@PiCoreTeam is looking into how stablecoins could expand its ecosystem, with a focus on payments, pricing, and bringing more users into its network.

The goal is to move beyond speculation and unlock more real world utility.

However, no specific timeline for a stablecoin integration has been announced yet.

Could stablecoins be the next major step for Pi Network?

[Tshepo Lekota @LkTshepo]

I came to skate and now I feel judged

[Tshepo Lekota @LkTshepo]

I get so nervous thinking about an agent with no instinct to walk away after a bad run

[Augustine @EyoAugusti73181]

i think one of the hardest things to build into a trading agent is knowing when to stop

after a few losses, the next entry can start looking like a chance to recover rather than a setup that genuinely meets the rules. that's where a strategy can drift away from the plan it was supposed to follow

the cooldown mechanism on @agenticscredit addresses that problem by pausing new entries after a losing streak instead of letting the agent keep firing

what i like about that approach is that it doesn't depend entirely on the agent continuing to make the right decision in a difficult moment. the risk rule is built into the system

that matters when every closed trade contributes to the ACS record. a bad streak is already useful information about how a strategy performs under pressure, and immediately adding more positions can make the damage harder to contain

paper trading helps build a record, but it doesn't guarantee live success, and a strong paper recovery can't simply cancel out weaker live performance

i wouldn't expect a cooldown to fix a bad strategy. i'd rather see it as a boundary that gives the system room to stop, reassess and avoid turning a losing run into an even larger drawdown

for autonomous trading, discipline shouldn't depend on whether the next setup looks tempting. the rules need to hold even when the results make breaking them feel attractive

@agenticscredit

[Tshepo Lekota @LkTshepo]

I need to know whether these agents actually diversify risk before I get impressed by the setup

[Ariel Lillie @ariell_xyz]

Five bots can still leave you with one big bet

Imagine several agents using different entry signals while all staying long BTC. Their names and trading rules differ, but a sharp sell-off could hurt them together.

That’s the portfolio question I bring to @agenticscredit multi-agent setup. Its whitepaper describes combining individual ACS ratings into a capital-weighted score for the operator.

Looking across the whole account makes sense. But a combined rating alone doesn’t tell me how much the strategies overlap.

I’d want to see shared positions, total directional exposure, and how often the agents lose at the same time. The published rollup description doesn’t establish how those relationships are measured.

Before adding another bot, I’d ask what risk it actually changes in the portfolio.

[Tshepo Lekota @LkTshepo]

Having to double check the network every time you move stablecoins is so annoying

[Tshepo Lekota @LkTshepo]

@Arifx0001 @utexocom @VIhnatiuk wait predictable fees for bitcoin payments is actually so compelling

[Tshepo Lekota @LkTshepo]

I need a full audit trail for every AI agent touching money because apparently I have trust issues

[Inference Labs @inference_labs]

Always-on agents create a new enterprise problem: AI can keep working even when nobody is actively watching.

Persistent systems need persistent evidence.

Inference Labs is focused on making autonomous activity easier to trace, verify, and audit.

[Tshepo Lekota @LkTshepo]

I love the idea of agents earning access to capital through a real trading history, this is so interesting

[Vexora ☣ @mdabdurrahim98]

An agent can have a strong trading record and still not have access to real capital. That is the gap Agentics Credit is trying to address.

Its @agenticscredit Score rates trading history on a 300–850 scale. The system looks at profitability, drawdown, consistency and other performance signals. A score of 580 is the qualifying threshold.

But qualifying does not mean capital is ready to deploy. According to the official docs, eligible agents still enter a funding waitlist. Real capital is not deployed on their behalf without an allocation.

There is another detail worth separating. Polymarket history can contribute to a score, but that does not mean the credit can be traded on Polymarket. The first approved execution venue is Avantis on Base.

The risk engine is also designed to enforce limits rather than simply warn the agent. Position sizes can be reduced after losses, and deeper drawdowns can trigger forced closures or halt trading.

So Agentics is building more than a score for autonomous traders. It is trying to connect a verifiable performance record with controlled access to capital.

The real test will be whether that score can lead to actual funding at scale. For now, clearing the bar proves eligibility, not access to a live credit line.

[Tshepo Lekota @LkTshepo]

The black pair is really cute and now I'm looking at Kai Cenat sneakers like this is normal

[Daily Noud @DailyNoud]

Kai Cenat has released brand new VIVET sneakers in two distinct colorways.

[Opti Max 💜🍀 @opti_maxx]

AI isn’t just improving the tools we already have.

It’s redefining what we consider possible.

For decades, technology was mostly about making humans faster.

AI is different.

It’s beginning to change how we think, create, analyze, communicate, build, invest, and interact with information itself.

The biggest shift isn’t that machines can now generate text, images, code, or videos.

It’s that intelligence is becoming increasingly accessible.

A small team can now operate like a much larger company.

An individual can research markets, build products, automate workflows, analyze massive datasets, and create things that previously required entire teams.

That changes the economics of creation.

It changes the economics of business.

And eventually, it changes the architecture of entire industries.

Finance is one of the areas where this convergence becomes especially interesting.

AI can process enormous amounts of information, identify patterns, automate decisions and make complex financial systems easier to interact with.

Blockchain brings another layer:

programmable ownership, transparent settlement, digital assets and global infrastructure.

Put those technologies together and the question becomes much bigger than:

“Can AI make our lives easier?”

The better question is:

What happens when intelligence becomes programmable and financial infrastructure becomes increasingly digital?

That’s where I think the next wave gets interesting.

We’re moving toward an internet where AI doesn’t just answer questions.

AI agents can research.

AI agents can execute.

AI agents can coordinate.

AI agents can interact with financial systems.

And blockchain can provide infrastructure for ownership, settlement and verifiable transactions.

We’re still extremely early.

The technology will evolve.

The infrastructure will improve.

The failures will teach us.

But one thing is becoming increasingly difficult to ignore:

AI isn’t simply another technological trend.

It’s becoming a new layer of the digital economy.

And the companies, protocols and builders figuring out how to connect AI with real-world industries could end up defining the next era of technology.

The question isn’t whether AI will change the world.

It’s who will build the systems that operate inside that new world.

That’s the part I’m watching. 👀

[Tshepo Lekota @LkTshepo]

The idea of an AI agent realizing it needs something and just paying a human to go get it is so good

[Crynyx @Web3_crynyx]

One part of @vangrid_io that I think deserves more attention is who can actually ask for physical-world data.

Traditionally, you’d expect a person or a team to realize they’re missing a capture, create a bounty, fund it, and wait for someone to do the work.

That assumption changes when the requester can be an AI agent.

An Agent can make the request itself, pay in USDC through x402 on Base or Arc, and initiate the job without needing an account or API key.

Think about what that enables.

A robotics system needs an updated view of a building.

A world model realizes it’s missing the interior of a particular site.

Instead of waiting for a human to notice the gap, the software can simply request the missing data.

But there’s an important reality check here.

AI can request the capture.

It still can’t walk into the building.

Someone has to visit the location, record the environment, and provide the raw observation. The reconstruction takes time too.

And honestly, that’s what makes the idea more interesting to me.

Vangrid isn’t trying to remove humans from the loop.

It is connecting machine demand with human physical presence.

The software knows what it needs.

A person can go get it.

The result becomes usable data for the system that requested it.

That creates a completely different way to think about physical-world data.

AI agents won’t only consume information from the internet.

They can eventually commission information from the real world.

And that is where @vangrid_io starts looking less like a mapping network and more like infrastructure for an AI economy that can actually reach outside the screen.

[Tshepo Lekota @LkTshepo]

Fitting rooms having one hook when you're trying on jeans and a top is so annoying

[delf | Cestra @delf002]

Stablecoins transfers on @SuiNetwork is free.