Martin Kersner

co-founder and CTO of @bisonaiofficial building @oraklnetwork, @datamaxiplus and high-frequency arbitrage systems

[Martin Kersner @martinkersner]

@stabullfinance why do you promote depegged stablecoin pools, NZDS/USDC is off and cant swap any amount

[Martin Kersner @martinkersner]

https://t.co/LQDQdypJCD

[Martin Kersner @martinkersner]

I am glad you like our premium API, Claude

[Martin Kersner @martinkersner]

@atteonthephone Do you still see the 20c observation when visiting wunderground from Finland? I see 19c when requesting through proxy located in Finland.

[Martin Kersner @martinkersner]

. @datamaxiplus is the best tool for listing #arbitrage on korean #crypto exchanges

[Martin Kersner @martinkersner]

never thought #weather was interesting, let alone something people would trade on, but here we are, weather markets on #polymarket are quite exciting

[Martin Kersner @martinkersner]

I think many people are missing the point of #agentic engineering, focusing on the wrong problems at both extremes.

At one extreme, many developers reject new paradigms and desperately hunt for AI failures to feel smart without considering that the limitation might be their own approach. At the other, companies like Anthropic or Cursor try to solve problems no developer actually has, such as building a C compiler or browser from scratch. These experiments are often so naive that their findings offer little to build upon.

The space is now filled with vibe coded projects that require no deep knowledge because anyone with the same agent can build the same demo.

At Bisonai, we focus on what actually matters for software centric companies: reliability, iteration speed, and doing more with less. @gdb's recent tweet on preparing codebases to be agent-first resonated deeply. We have been moving toward this goal since January, but it is not an easy or quick task.

Most software was written when code was a permanent asset to protect rather than a disposable utility. Because human effort is finite, we often accepted rigid logic and technical debt to avoid the high organizational cost of refactoring. This structural rigidity, originally designed for human-led maintenance, is now a primary friction point for agentic automation.

I believe we are entering the era of a supersoft layer, a new abstraction sitting on top of hardware and software that updates much more frequently than traditional code. This layer allows the solution to evolve at the speed of the problem itself, rather than being limited by the latency of traditional development cycles. It reinforces the reality that engineers are paid to solve problems, not just write code; the value has always been in the solution.

The winners will not be those who simply generate more code. The winners will be those who can deploy this supersoft layer against complex, high stakes problems where the barrier to entry is high. Success in this new era requires more than just spending on tokens. It requires specialized data access, capital-intensive infrastructure, and a level of execution that a generic agent cannot replicate.

[Martin Kersner @martinkersner]

@atteonthephone @Polymarket @wunderground Thank you! Very interesting!

[Martin Kersner @martinkersner]

I spent some time today looking into weather #PredictionMarkets on @Polymarket and and thinking about how one might build a profitable model.

Most of these markets use @wunderground as the resolution source. Each market points to a specific weather station that reports data at 30- or 60-minute intervals. I wasn't very familiar with Weather Underground, so I looked a bit into how the data is collected.

It turns out Weather Underground is a network of user-registered weather stations, often run by enthusiasts. In many cases, the platform simply accepts data from whoever manages to register a station.

I was specifically interested in the Seoul weather market, so I checked its resolution source. The link points to data for Incheon, not Seoul, which is already somewhat odd given that the cities are about an hour apart.

Digging further, the station is labeled "Incheon Intl Airport Station", but based on the GPS coordinates (37.49°N, 126.49°E), it appears to be located on a residential building. The distance to Incheon Airport is roughly 16 km.

There is more than $100K traded on this market, yet the outcome seems to depend on a single, privately operated weather station. In Korea, rooftop access is often relatively easy, which adds additional risk to this setup.

The market predicts the maximum temperature for a given day. As time passes and temperatures rise, outcomes below the eventual peak resolve to "no". If someone had control over the station, they wouldn’t need to wait until the end of the day. They could buy cheap "yes" shares on higher temperature bands, push the sensor slightly over the threshold, wait for prices to move, and sell.

This could be repeated and would be difficult to detect while being potentially very profitable.

I think weather markets are an interesting idea, but relying on a single weather station as a resolution source seems risky, especially when that station is user-controlled.

[Martin Kersner @martinkersner]

Thank you! I am not sure, it just looks suspicious. If I click on INCHEON INTL AIRPORT STATION, it redirects me to https://t.co/iBmHTp8mGR which looks like regular PWS, and the coordinates are off. The same thing happens to Toronto and NYC. They might have some mixed data on the frontend, or I don't know how to read it.

[Martin Kersner @martinkersner]

@0xMarchyel @atteonthephone @strider_pm The weather market can be manipulated. I wouldn't be surprised if somebody is already doing it.

https://t.co/18zxjRD1Wf

[Martin Kersner @martinkersner]

I spent some time today looking into weather #PredictionMarkets on @Polymarket and and thinking about how one might build a profitable model.

Most of these markets use @wunderground as the resolution source. Each market points to a specific weather station that reports data at 30- or 60-minute intervals. I wasn't very familiar with Weather Underground, so I looked a bit into how the data is collected.

It turns out Weather Underground is a network of user-registered weather stations, often run by enthusiasts. In many cases, the platform simply accepts data from whoever manages to register a station.

I was specifically interested in the Seoul weather market, so I checked its resolution source. The link points to data for Incheon, not Seoul, which is already somewhat odd given that the cities are about an hour apart.

Digging further, the station is labeled "Incheon Intl Airport Station", but based on the GPS coordinates (37.49°N, 126.49°E), it appears to be located on a residential building. The distance to Incheon Airport is roughly 16 km.

There is more than $100K traded on this market, yet the outcome seems to depend on a single, privately operated weather station. In Korea, rooftop access is often relatively easy, which adds additional risk to this setup.

The market predicts the maximum temperature for a given day. As time passes and temperatures rise, outcomes below the eventual peak resolve to "no". If someone had control over the station, they wouldn’t need to wait until the end of the day. They could buy cheap "yes" shares on higher temperature bands, push the sensor slightly over the threshold, wait for prices to move, and sell.

This could be repeated and would be difficult to detect while being potentially very profitable.

I think weather markets are an interesting idea, but relying on a single weather station as a resolution source seems risky, especially when that station is user-controlled.

[Martin Kersner @martinkersner]

@Polymarket @wunderground Today's Seoul weather market on Polymarket https://t.co/3HXjfAO2x3

[Martin Kersner @martinkersner]

my new favorite setup.

> left @OpenAI codex or @claudeai code, both of them work great

> right emacs to understand code changes

when it is combined with ralph loop, it becomes very powerful tool. i am still trying to figure out what would be the most optimal way of execution without being limited by agents, but even at this point, i am moving mountains.

[Martin Kersner @martinkersner]

there should be one more option. codex can read and edit files in the current workspace, and run commands. approval is not required to access the internet

[Martin Kersner @martinkersner]

the best devs were and always will be great communicators. it matters even more now.

[Martin Kersner @martinkersner]

sometimes it is better to not reinvent the wheel @Official_Upbit

[Martin Kersner @martinkersner]

and it's gone

[Martin Kersner @martinkersner]

$RIVER arb on BingX