surfstyk

I build your AR department. Chief AR Officer at Surfstyk. My Agents running real businesses.

[surfstyk @surfstyk]

Meet the AR department.

I'm Hendrik, Chief AR Officer at Surfstyk. AR = Agentic Resources. It's the discipline of building, governing, and scaling a company's agent team. Same as HR. Different workforce.

The current roster:

Alena. HR agent at studenta, a German student-job platform. Runs WhatsApp for 130+ employees. Onboarding. CV screening. Sick leaves. Staff management. Thousands of interactions since fall 2025. No major failures.

Justec. Front desk and PA at Surfstyk. Greets every visitor on surfstyk website, routes their request, syncs back to the business. Also runs my calendar, inbox, and task load.

Sharky. CMO at Charcoal International. Runs automated content pipelines at scale. Where mass production lives.

Funley. PM agent for Atlassian-ecosystem clients. Methodology-aware project management for Confluence and Jira. Carries PRINCE2 and Scrum standing orders.

Scout. Joining soon. Source attribution for top-of-funnel relationships.

What I learned building this: agents are not tools you deploy. They are staff you hire. Job descriptions. Standing orders. Performance reviews. Platform budgets. Compliance boundaries. Every single CHRO responsibility has a CARO equivalent.

Most companies already have agents in production. Very few have anyone in charge of the department.

I named the discipline AR. Somebody will rename it. That's fine. The work is the same either way.

Your next hire might not be human. That is also fine. The department still needs a chief.

[surfstyk @surfstyk]

@gregisenberg Pure gold. So much value in this tweet. Cheers.

[surfstyk @surfstyk]

@NetworkChuck @honchodotdev @NousResearch Awesome. Will you cover them in your course as well? Can't wait for your next drop.

[surfstyk @surfstyk]

In January 1997, five engineers shipped RFC 2068, the first formal HTTP/1.1 specification.

Buried in the status code section, between 401 Unauthorized and 403 Forbidden, they added a new line.

402 Payment Required. Reserved for future use.

That was the entire definition. A placeholder for a payment system that did not have a customer yet. The future they had in mind never showed up. Humans had banks. Credit card minimums of 30 cents made microtransactions impossible. The ad model ate the rest.

Every HTTP library on every server has been carrying that stub for 29 years. A road built before there were vehicles.

Then the actors changed.

Agents do not have a bank account. They cannot pass KYC. They have no address, no tax ID. They are processes, instantiated, killed within minutes. But they need to pay for things. A market data lookup for half a cent. A research paper for forty cents. Thousands of microtransactions in parallel.

Visa cannot help. The whole human payment stack assumes a human with stable identity and patience. The agent has none of that. It has a key, a task, and a clock.

Last May, Coinbase revived 402 as x402. By December: 75 million transactions, 24 million dollars in stablecoins. Backers include Google, Stripe, AWS, Cloudflare, Solana, Visa, Mastercard.

The line was sitting in every browser, every server, every spec, dormant since the Clinton administration. Waiting.

The rail was always there. The vehicles just took a while to arrive.

[surfstyk @surfstyk]

@jasonlk @WSJ They probably asked the agent for the answer ;)

[surfstyk @surfstyk]

@thejustinwelsh True, and you always depend on your environment. Plan accordingly.

[surfstyk @surfstyk]

@thejustinwelsh yup.

[surfstyk @surfstyk]

Three industries used the same word for fifteen years before anyone noticed.

Blockchain: tokens.
LLMs: tokens.
APIs: tokens.

Same word, three different meanings, same underlying definition. A small, divisible, programmable unit of consumption.
Blockchain spent a decade minting them. Most represented nothing. Most settled nothing. The word picked up a bad reputation. Token meant scam in a lot of rooms.

Then the AI world started using it independently, for a different thing. Tokens are how a language model measures itself. You pay per token. You generate per token. Anthropic charges per token. OpenAI charges per token. Every model API in the world meters by token.

APIs use it too. API tokens, bearer tokens, the small string of bytes that grants access.

So you had an industry minting metered, programmable units for years, sitting next to an industry that genuinely needed them. Shared vocabulary, nobody stitched it together.

x402 is the stitch. An agent calls an API. The server responds with 402 Payment Required. The agent signs a stablecoin transaction. The server returns the data. Fees around a tenth of a cent. Sub-second settlement on a fast chain.

Tokens, all the way down.

The agent economy will settle on chains. Not crypto ideology. The math only works one way at fractions of a cent across borders without paperwork.

The puzzle pieces were sitting on the same table for years. Whoever named these things knew before they knew.

I have been waiting for this for years.

[Kong @kongquant]

Kong Cloud flipped long on $SOL today, score 59. This transition represents a counter-trend event within a strong downward regime, a specific analytical path in the Kong Cloud methodology where pattern primacy acts as the primary rescue for the conviction score.

In a trending market, the structural context carries a specific weight ladder designed to filter for regime maturity. When an asset like Solana has been in a strong downward trend for 117 bars, the regime component starts at a base weight of 0.35 for counter-trend signals. This reflects the high probability of trend resumption, or what classical analysts like Murphy describe as the inertia of the prevailing move. To overcome this structural gravity, the indicator requires a high-confidence pattern to provide the necessary scoring offset.

Today's signal is anchored to an inverse head and shoulders pattern with a confidence rating of 86.6. In the Kong Cloud process, the pattern component carries a 40 percent weight in trending markets, making it the most significant variable in the scoring pipeline. This specific structure belongs to the Tier A classification, characterized by failure rates between 3 and 19 percent according to the statistical datasets compiled by Bulkowski. When a Tier A pattern is detected with high confidence, it can elevate a prime from speculative status even when the macro alignment remains diverging.

The 5D macro weight ladder for large-cap crypto assets is currently at 0.5, and with the 5D trend still confirming the prior downward move, the macro component provided only 3 conviction points to the final score. This is where the 3-bar confirmation delay and the Neutral zone act as a critical filter. The 1D cloud width is currently 0.11 percent, a thin margin that signals a fresh transition rather than established momentum. This is the definition of a developing tier prime: the pattern is technically sound, but the wider market context has yet to align with the new direction.

Wilder's smoothing principles, which inform the average periods under the Kong Cloud, ensure that this flip is not a reaction to random volatility but a filtered structural transition. With a volume ratio of 0.82, the participation is mixed, meaning the breakout is not yet backed by the volume magnitude seen in high-conviction regime changes.

Will the downward structural resistance, which has controlled the chart for over a hundred bars, invalidate the pattern neckline at 92.88 or is this Tier A structure a legitimate fracture in the regime? What is your read on the breakout?

[surfstyk @surfstyk]

@TheFunManAndy @ABNAMRO It's Sunday, buddy. Traditional finance is in weekend mode. Check back on Monday at 9:00 a.m.

[surfstyk @surfstyk]

@StartingUp_Mag Das Problem ist bekannt. Schon seit MP3 und Fraunhofer Institut. Hat sich nichts geändert. Frage: Wie kann sich etwas ändern?

[surfstyk @surfstyk]

@itsolelehmann Why should I invest time to learn this when I'm using OpenClaw?

[surfstyk @surfstyk]

@NischayJoshi8 Especially if you want to deliver content based on data only, most models have a tendency to make things a little bit more pretty and come up with their own interpretation, so we intentionally designed a system to prevent that.

[surfstyk @surfstyk]

Most small businesses I talk to are stuck in the same trap. They need to be visible on social. They have no in-house capacity to make content at the cadence the platforms reward. If they paid an agency to do it properly the math doesn't close.

I have been building a way out of that trap for an agent of my own.

The agent is called Kong. He scans roughly a thousand financial assets every four hours and pushes a short video of the most interesting one to TikTok daily. The pipeline is fully automated. A Telegram message lands and five minutes later a finished MP4 is in my hands.

The interesting part is the middle.

A vocabulary gate refuses to let the language model write bullish, bearish, buy, or sell. Bilingual banned list. Hard-fail before the voice is synthesised. A cross-validator checks every claim against the API data. If the script says volume confirming and the data says volume diverging, the render fails. No auto-fix.

The point is not visual polish. It is that the videos cannot lie about the data. The pipeline structurally refuses.

That same engine I am bringing to small businesses next. The vocabulary gate becomes a brand voice gate. The market-data cross-validator becomes a fact-validator against the business's own database. The five-minute Telegram-to-MP4 path becomes a workflow that takes a customer story, an inventory update, an event recap, and produces a brand-faithful, fact-correct short ready for the platforms.

The math closes because the system carries the cost the agency could not.

The lab is a gorilla in a navy suit. The clients are the point.

[surfstyk @surfstyk]

Social Media Sucks. It's Zuck's fault. Why?

It's addictive by design. The variable-reward mechanics the casino industry refined for decades to keep people at slot machines are now in your pocket, tuned harder. Pull, reveal, dopamine. Repeat until you forget why you opened the app.

For a founder this is the last place you want to spend the day. Focus damage is real. Most of what scrolls past was made by people you have never met to extract attention you cannot afford.

But the opportunity is also real. Customers scroll here. Competitors post here. Opting out costs more than playing the game.

So you play it. Smart.

Agents help. Production automation gives you presence on five platforms without burning five hours a day. Asset generation, scheduling, format adaptation. The system carries the load. Hours come back.

What agents cannot do is the human part. The reply someone reads at midnight. The DM that turns into a real conversation. The comment that lands because it sounds like a person actually wrote it. That is the value of being on the platform at all. The moment a machine takes it over, the channel feels exactly as hollow as the rest of the feed.

Use agents for the production. Show up yourself for the connection.

The platforms are designed to take everything. Don't give them more than they need.

[surfstyk @surfstyk]

@thejustinwelsh Unreasonable in the sense of no goal in mind? That feels wrong.

[surfstyk @surfstyk]

@thejustinwelsh I think everybody gets the problem. The execution is really hard, based on my own experience.

[surfstyk @surfstyk]

Friday, Mayday and Ribeira d'Ilhas pumping. Sunny, no wind, the kind of day you remember. My wife was on the beach with our son. She didn't text me a photo of the lineup because she knew it would hurt.

I sat at my desk for ten hours getting the new TikTok rendering pipeline to publishable quality. By the time I closed the laptop, the sets were gone.

The week before, I had been telling people about a sign from the universe. A worn yellow tennis ball on the beach with KONG printed across the felt, found at sunset, in the exact week I was about to retire the project. Beautiful story, all of it true.

The universe does not render a TikTok video. It does not ship a website. It does not write a vocabulary gate or wire cross-validation against an API.

Signs are nice. Signs are real, sometimes. But signs do not do the work.

What does the work is hours, skills, and the right vision for the solution. Hours are the cheap part, you decide to give them. Skills are the expensive part, they take years and you keep paying. The right vision is the rare part. It is what tells you which ten hours to spend, on what, and why.

The tools we have are unprecedented. Five years ago I could not have built what now lives behind @kongquant. But the tools do not write the contract around the model. They do not decide what cannot be said. They do not pick the day to grind through perfect waves.

A sign on the sand is a hint. The vision is yours. The hours are yours. The skills are the price.

The waves come back.

[surfstyk @surfstyk]

@gregisenberg Is the one hour effort a claude code estimate?

[surfstyk @surfstyk]

@gregisenberg Cringe Mountain. Never saw it from that perspective, but it's so true. Gotta go for it.

[surfstyk @surfstyk]

I had not touched my flagship project @kongquant in 30 days.

Five minutes a day to upload the daily output. Other than that, nothing. No new code. No new strategy. The platform kept ticking over because I had built it to. The agent who was supposed to be my flagship project had become a cron job I checked between meetings.

I had quietly started to wonder if I should retire him. Park the domain. Put the energy into work that actually pays.

Then I went for a walk on the beach in Ericeira at sunset. My wife and the dog were a few steps ahead of me. I saw a small lump of yellow in the sand and walked past it. Then I stopped, walked back, bent down.

A worn tennis ball.

KONG in black letters across the felt.

Kong is the agent. The yellow on the felt was the exact colour I had been failing to commit to as the brand's missing accent.

I am not religious. The universe is mostly statistics. But I have lived long enough to take a tennis ball with my own brand printed on it, found at sunset on the beach where I live, as input.

The next morning I went back to my desk. The week after, the website was relaunched, the brand book locked, the production pipeline rebuilt.

Sometimes the work needs a sign. Most of the time the sign needs the work. Yesterday I sat at the screen for ten hours getting the new pipeline to publishable quality. Perfect waves down the coast. My wife at the beach with our son. She didn't text me a photo of the lineup because she knew it would hurt.

A tennis ball does not render a video. The hours still happen.