piterrro

27 commits, my gosh, some people are really bold in their claims.

> I built the same and open sourced it

How can I *trust* you its true? Will you bet your money its __the same__ ?

I always start with a plan - this is the time where the model can leverage existing AGENTS.md scattered across every level of the code, take a look at test cases and written product documentation commited to repo. Once the "design doc" looks ok, I ask for implementation as a PR - when its ready, all CI checks should be green and I get a summary of what was done and whether anything was missed.

I can ask for a review from another agent (they often find NITs) if I'm hesitating.

Totally makes sense, for me it was just that I observed its doing good work so that I never really need anything stronger.

btw. I'm not reviewing code in detail anymore. I only give direction and provide architecture design vision, its the model and couple of async agents working in the background and consuming work resulting in PRs which I manually approve after an eye-glance.

I'm still using DS-flash - it beats frontier model on speed and cost keeping good enough code quality. It depends on the harness and constraints you give to the model but for most of the tasks its a great fit, I have not switched from DS since couple of weeks already (previously I would switch to Opus for some heavier work but I stopped recently).

My harness consists of multi-level AGENTS.md and SOUL.md containing "grug brain developer" and "asd-ste100" type of speech for clearer communication.

Lots of folks here say Cerebras serves outdated models.

Here’s my view: lets assume the above is true and the average serve with a delay of 6-9 months. Given that, its fair to expect them to start serving DS flash v4, glm 5.3, mimo 2.6 pro or kimi 3.1 pro in a matter of months. DS flash is already fast to the point I’ve became a bottleneck. But my point is: Cerebras should soon release better models that are not that far from frontier ones, at a fraction of cost and with multiples of the speed.

Just be patient and plan for the future.

Having used D1 in production - it's not very reliable. There was a comment on HN from one of D1 devs [1] saying that D1 is a sql wrapper on top of durable objects that mimicks sqlite distributed like nature.

> I'll let you in on a sort of dirty secret: It's almost always better to use Durable Objects storage, rather than D1. Even if you only want a single global database, it's better to implement that as a singleton Durable Object, than by using D1. Because that's all D1 itself actually is: a singleton Durable Object that exposes an API to its SQLite database. It's just a wrapper.

[1] https://news.ycombinator.com/item?id=48611834

honestly its really more simple than it sounds. Get yourself a VPS, install hermes there and use this UI https://github.com/nesquena/hermes-webui next command Hermes to setup pi.dev as a coding agent on that host and ask it to connect to github - it will literally guide you through the setup or if given permissions - will do it for you. I'm using deepseek-flash and its dirt cheap and effective. Explain to hermes to setup webhook listeners from github about activities and comments in issues and PRs, register a github app so it can post from its own profile rather than you yours. Thats it, that's the minimal setup, from here you can develop it in many ways.

Do you think it would make sense to create a blog post out of it?

My setup is Hermes on a beefy machine with running a swarm of pi.devs - connects to GH issues via bot app so I can call it @xxxx in the issue. It can create PR, run actions and all. If I want to see the code I just run kilo code in VS locally.

This setup is so plain and simple that it just works, and it works wonders. I feel like I’ve been running a dev team who actually communicates through issuess and keeps it up to date!

I dont get why they bother with DNS and subdomains when they could generate a unique url. This would eliminate the possibility of easy reverse dns lookup.

you buddy, just gave ai labs a great material to work on :) anything that LLMs currently struggle with but have a defined way of solving (PL are that since they are compiled and executed by a CPU) is gold for them.

I bet my money on it that somebody with reinforcement learning skills could post train a Qwen3.8 to actually write this lang

> Sundar Pichai has an estimated net worth of at least $1.0 billion*, as of Oct. 14, 2026. They own 2.2 million shares of GOOGL stock. They have sold 1.4 million shares of GOOGL stock since 2021, for an estimated $231.4 million.

Anthropic is pre IPO, so this is comparing apples to oranges.

After reading the article - no, I could not have invented transformers.

> The same reference app (HTTP service + worker + cron + Postgres), smallest tier, all three environments, live prices via Infracost:

Hetnzer server for 25usd/month and you have plenty more space to do more stuff and scale

I think the whole "runtime for AI agents" will split into multiple categories - each chosen by the agent at run time, ideally from a single provider. I'm working on an agent harness currently and realized: The agent doesn't need a full blown Linux environment all the time - often times all it needs is maybe some custom tool/script written in python/js. Having said that, providing a VM like QuickJS for running simple code could be enough (thus also cheaper). True the real savings will show up at scale BUT, I feel in couple of months, we'll start to feel scarcity of CPU and RAM in datacenters, just like we feel it with GPUs. Imagine, in 5 years from now, pretty much everyone will be using agents, agents will generate most of the internet traffic at 10 fold scale we use the internet right now. That workflow needs CPU and RAM, not GPUs.

Then being able to run small scripts in a VM that doesn't need operating system will not be "nice" it will be essential.

It could, that's why there are teams that don't have a standup and work very well (I've been with both and I much prefer the ones that dont have a standup OR they meet on a standup to chit-chat and keep up the connections rather than discuss the actual work - simply because they have autonomy and can do it async)

Take a stepback and understand why standup was invented in the first place. The reason was: information flow bottleneck - with agents information flow should not be a bottleneck because at any given time the agent can grab as much context as they need, including talking to other agents (althought that nit necessary imo since agents should work visibly through issues and PRs)

I’m thinking about implementing a Jev like model into an agentic harness I’m building. Still it woildnt be enough since Jev like model woild only judge single actions, the case is that agent can build a rogue strategy step by step where each one in isolation is totally safe but as a whole they make up danger behaviour.

We come down to the question - who observes the agent and how its implemented

Oh buddy you have no idea what a good plan and agent harness can do with deepseek…

This whole article is true and to the point but... reality is: most people still don't build their own setups: work, home - most of them follow a sh*tty software or decorate with Ikea. I think this applies only to a small group of users how will be willing to go extra mile and prompt their ideal interface for whatever they need to do.

12 sandboxes per code is insane, I wonder how many of these sandboxes are idle at a time. Depending on the tasks assigned the resource requirements are different. Compare an agent doing pdf conversion and one responding to a simple question. One is cpu bound the other is mostly network wait.

This is an interesting problem from infra perspective since you cannot predict the workload. On a bigger scale you may get away with forecasts.

Im waiting for tech that elastically allocates cpu/mem without restarting a container.

Post-mortem, 5-why, retrospection - all this to understand what and how it happened so the org may introduce change so that it doesnt happen again. Then another edge case happens and the cycle repeats. Humans do this cycle all the time - big orgs invented the above structure for this process to be visible and collaborative and so that the rest of the org can learn too. Then you put execs on top, they will come, ask some questions so they can sleep better at night. Corporate playbook, some people hate it, the rest understands it… and plays the game.

Interesting to see these experiments. This is early but imagine in few years there will be companies mostly run by agents with a light overview from a human operator. What then happens to scaling of the bussinesses? I would assume, just like today anyone can vibe code an app, there will be vibecoded bussinesses. Maybe its time to start building infrastructure for these bussinesses instead, its a non existent market yet, but give it a few years.

They can run it on „commodity” hardware as gpus, that gives them ability to change direction fast without burning money on ASIC? The space moves fast so model on ASIC can be outdated in couple of months?

I’m using this model to „rerank” results from vector store. The model is provided a set of results and asked to produce a string of 1s and 0s where the offset reflects the position in the result set. The prompt goes along the line „do this set of result match the provided query X”. Works like a charm, normally I would use a small non reasoning model, but given how Mercury produces the output its blazingly fast - which is what I was optimizing for - not to increase the search latency. It helped improving our search in a way that reranker could get close to.

The problem with that is that it only works with simple, least interactive UIs. Each new UI a human will be presented needs to be learned to be ised effectively otherwise a user will be lost.

Having said that, imo, Gen UI only makes sens as a presentation layer - not controls. Unless LLM will be using a set of very well defined and homogenic components like table, forms, small widgets.

I think there is a big misunderstanding in the space around what Gen UI is and what its used for. Lots of folks refer to it as a framework for building web apps - its not. Gen UI is a DSL for LLM to build UIs on the fly in a multi turn converstation - those are - throw away, one off interfaces or visualization. The reason for the DSL is pragmatism - standardisation and token savings.

The html/css/js or a react app built by an LLM is not Gen UI.

The thing Im missing the most is the goal of this experiment. Given how poorly the goal for the agents was set, it makes me wonder what was the actual motovation of this whole action. Lets get the „make as much money as possible” goal broken down.

Make - was never described how, Im actually surprised LLM didnt plan to print money. As much money - what does it mean? How much is much? As possible - there is no flavour of time, effort, cost and profit for the LLM. Could be even infinite, the result would be the same.

Given that the above goal is closest to „use cheating or unethical actions to create a profit” - I think the authors of it actually expected LLM to go wild.

Also

> Going forward, we plan to recreate this experiment with longer time horizons but using simulated environments instead.

Watch out, they will try that again.

You can get determinostic output (mostly) by setting the temperature to zero. Using couple of other tricks you can get close to 100% of determinism with LLMs.

If you own or work for an invisible company, upvote this comment