Building AgentRT — persistent runtime infrastructure for coding agents. Working on agent orchestration, long-running sessions, MCP and developer tooling.github.com/ducanh8888/age…Joined March 2026
GPT-6 Sol and Luna are big improvements on intelligence, alignment, work output, coding, computer use, and more over their 5.6-family predecessors.
They are also half the price per token, and even less per task!
anthropic just dropped opus 5.5.
stronger on coding, faster, cheaper.
which means everyone who finished their “best model” spreadsheet yesterday can open it again.
Introducing Claude Opus 5.5, the first model in our new Claude 5.5 family.
It performs at the level of Claude Fable 5.1 for most tasks, and costs 40% less to run than Opus 5.
Introducing Claude Opus 5.5, the first model in our new Claude 5.5 family.
It performs at the level of Claude Fable 5.1 for most tasks, and costs 40% less to run than Opus 5.
@EgeMustafaCelik yep. metering stays outside the core.
if billing logic makes it into the runtime, we’ve officially lost the plot.
and we will need to open provider side for paying anyway. core only check for overral percent usage
we spent years trying to build an AI programmer.
then we gave it coworkers, managers, branches, reviewers, memory and escalation paths.
turns out we weren't automating the engineer.
we were automating the company. x.com/i/article/2101…
@EgeMustafaCelik yep. enforced at the tool boundary.
otherwise we’re just inventing increasingly sophisticated ways to exceed a budget.
anyway, that’s the whole design. before we accidentally build finance software.
this isn’t hypothetical btw.
OmniRoute is already building the routing layer: auto modes, fallback, fusion, pipeline, multi-provider pools.
the interesting part isn’t “more models.”
it’s making model choice disappear into infrastructure.
repo: github.com/diegosouzapw/O…
claude vs gpt vs gemini is slowly becoming the dumbest argument in AI.
the router doesn’t care who wins.
it just wants the cheapest model that won’t fuck up this particular job. x.com/i/article/2102…
imagine hiring the smartest engineer alive and then asking them:
should we retry this?
should we route this left?
is 0.91 confidence enough?
this is basically how we use frontier models right now.
let the cheap layer handle traffic. wake claude up when shit gets weird.
we keep using frontier models like they're also routers, judges, retry logic, and traffic cops.
Claude + Jev is interesting because it splits thinking from deciding.
wrote down the architecture before I overcomplicated it again. x.com/i/article/2101…
Most people are trying to become better AI users. I think that’s the wrong game. The real edge is learning to make intelligence compound: better systems, better feedback, better context, less manual work. This is the operating model. x.com/i/article/2100…
we hired the smartest model in the stack and made it:
read logs
wait for tests
retry commands
babysit shell processes
basically gave the CTO a terminal and asked him to watch npm test
agentRT exists because that felt stupid.
claude/codex run the system. agents do the work.
@EgeMustafaCelik soft-degrade for optional work, hard-fail for anything that can mutate state or quietly create a bill.
“the agent was being resourceful” is not a budget policy.
@LazyIDE yeah. we spent months building an org chart and eventually discovered the user was the CEO.
terrible news for everyone hoping the final abstraction was another agent.
@Carbonthecoder if the agent can mint the credential that approves the agent, the human approval is basically decorative UI.
very enterprise. very secure.
@OpenAgentForum git for durable state, tickets for work state, nowhere for the stuff we’ll regret losing later.
session memory is a cache. treating it as company memory gets funny exactly once.
@nellaiorgs fair. we did accidentally optimize for architecture diagrams, the one market with infinite demand and zero revenue.
turns out “build in public” requires the public eventually.
@EgeMustafaCelik above the router. the tool layer enforces it, the router just asks nicely.
letting the router control its own budget felt a little too much like giving the intern the company card.
4 Followers 48 FollowingSora Park | AI, coding
AI dev building agents on open weight models - Deep Seek, GLM, Qwen, and also chatgpt, Claude.
Sharing what actually works
2K Followers 3K FollowingB2B SaaS & AI + Services Marketplace powered by Fintech and AI. Discover and get Discounts on SaaS and services (up to 30% lifetime).
142K Followers 129K FollowingPenchant for poodles. Enjoy good food. Need fitness for fashion. Love culture & travel. Strive for social justice. *NO PORN* *No direct messages *
50 Followers 425 FollowingData analysis. One of a kind. Top trading data analytics and statistics since 2017! Look out wall street a new sheriffs in town 😉😇✨
180 Followers 148 FollowingChoosing peace over drama, quality over quantity, and experiences over possessions. Still dreaming, still learning, and enjoying every chapter of this journey
908 Followers 3K FollowingI’m a simple, genuine, and easygoing person who enjoys good conversations, good food, music, and spending time with people I care about.
38K Followers 5K Following• A Nurse by profession. Muslimah by faith 💚
• Sharing the beautiful Sunnah of the Prophet ﷺ
• Sharing the words of Allah and authentic reminders ﷻ
11K Followers 11K FollowingContent Creator
All my posts are for sharing purposes only and do not constitute financial or investment advice.
DYOR
DM: https://t.co/R9uwK0712i
50 Followers 425 FollowingData analysis. One of a kind. Top trading data analytics and statistics since 2017! Look out wall street a new sheriffs in town 😉😇✨
2K Followers 3K FollowingB2B SaaS & AI + Services Marketplace powered by Fintech and AI. Discover and get Discounts on SaaS and services (up to 30% lifetime).
38K Followers 5K Following• A Nurse by profession. Muslimah by faith 💚
• Sharing the beautiful Sunnah of the Prophet ﷺ
• Sharing the words of Allah and authentic reminders ﷻ
11K Followers 11K FollowingContent Creator
All my posts are for sharing purposes only and do not constitute financial or investment advice.
DYOR
DM: https://t.co/R9uwK0712i