elvis(@omarsar0)

Start with something fun, like a fake Crocs x Porsche campaign. You can try it here: https://t.co/eh...

elvis (@omarsar0) — Wed, 07 Oct 2026 20:46:38 GMT

Start with something fun, like a fake Crocs x Porsche campaign. You can try it here: elements.envato.com/lp/burst-mode/…)

💬1🔄0❤️3👀1329📊2

AI image tools have gotten very good at giving you exactly what you ask for. Knowing what to ask f...

elvis (@omarsar0) — Wed, 07 Oct 2026 20:46:37 GMT

AI image tools have gotten very good at giving you exactly what you ask for. 

Knowing what to ask for is still hard.

@envato just launched Burst Mode to help. 

You start from a rough idea or a reference image, get up to 6 directions side by side, then steer them with moodboards and a creativity control that goes from focused to wild.

A really good experience and a great way to get strong results fast.

Envato@envato

Most AI tools expect you to know exactly what you want.

But creative work often starts with a brief, a reference, or just a hunch.

That’s why we built Burst mode.

Turn the start of an idea into up to six image directions at once, up to 10× faster, for just one AI credit.

More directions to explore. More chances to find the one worth pursuing. elements.envato.com/lp/burst-mode/…

Your browser does not support the video tag.

💬8🔄0❤️16👀7107📊7

Own your intelligence stack, folks! This chart could mean so many different things without raw deta...

elvis (@omarsar0) — Wed, 07 Oct 2026 17:28:35 GMT

Own your intelligence stack, folks!

This chart could mean so many different things without raw details.

I suspect more companies/devs are starting to realize the business opportunities RL unlocks, given the impressive models we already have today. 

For many real-world tasks, you don't need AGI; you need a proper specialized model, harness, and data flywheel. 

A new, fruitful post-training era is upon us. It's hard to see, but it's starting to happen with full-stack AI companies, and it's only going to keep growing.

Vincent Weisser@vincentweisser

Post-training / RL and inference is taking over compute

💬11🔄4❤️38👀8392📊14

Agent-to-agent communication with a personal agent on top is wild. Wow. I underestimated how well ...

elvis (@omarsar0) — Wed, 07 Oct 2026 17:07:47 GMT

Agent-to-agent communication with a personal agent on top is wild. 

Wow. I underestimated how well models like Opus 5.5 can coordinate already. 

This opens a whole new set of experiences with agents.

But it takes a bit of tuning (automations, webhooks, system prompts, tooling, etc.) and retrieving context at the right time. 

I don't have a perfect setup yet, but I feel like I'm already living in the future. 

My agent team can do work (higher quality than I could imagine) at speeds I can't even keep up with. 

It's crazy how fast things have moved in about a year. 

I started with individual Claude Code sessions.

Then, just a couple of months later, I moved to multi-agent systems with subagents, parallelizing work. 

Then, a couple of months ago, I built a persistent, higher-level agent team, similar to Grok Bot. Of course, with my own agent orchestrator and a team of like 8 specialized bots (CFO, CSO, CTO, CMO, CCO, CoS, etc.)

Then, about a month ago, I started playing around with agent-to-agent communication with a personal agent. 

And I can tell you, most of the apps like Code and Claude Desktop are way behind. But they will catch up.

This is why I believe we should all be building our own agent orchestrator/harness. You need to personalize to what works for you, which creates an important edge. You can start with a fork; who cares. Start somewhere. 

It's just mind-blowing to see agents coordinate and do things on their own that would take a village or an entire org in some cases. 

Just sharing my raw thoughts here, but things will dramatically change again in the next couple of months, leading into next year. 

Sharing a snapshot of one of my personal agents, Clio (Chief Content Officer), doing research all day and putting together high-quality content to inform me on things I am interested in and things I am writing about on X.

💬37🔄11❤️91👀9850📊59

Get started here: https://t.co/gjhucdHXdA

elvis (@omarsar0) — Wed, 07 Oct 2026 15:37:41 GMT

Get started here: ref.viktor.com/elvis-x-6

💬0🔄0❤️8👀1310📊2

Reading eval results is now the slowest part of building agents. I'm Elvis, founder of @dair_ai. I ...

elvis (@omarsar0) — Wed, 07 Oct 2026 15:37:24 GMT

Reading eval results is now the slowest part of building agents.

I'm Elvis, founder of @dair_ai. I lead research, build, and teach about AI agents.

I run harness experiments every night, but reading the results was eating my mornings.

Every change to my harness gets evaluated overnight, whether it touches memory, tool use, or context compaction.

The morning after is the hard part. I check which tasks my agent got right yesterday but wrong today. Then I open the logs for each failure, one by one, to figure out which of my changes caused it.

I tried a dashboard first. It showed the pass rate dropped. It couldn't tell me why.

That is the job Viktor, an AI employee in Slack, is built for. He reviews the results overnight.

Here is how that plays out. Say 23 tasks that passed yesterday fail today. Viktor checks all 23 logs, traces them to the one change that caused them, and suggests undoing it. I check the logs and make the call.

Viktor does the digging. I decide what goes into the harness.

He is also proactive. He flags problems before you ask, which helps you stay on track with complex eval runs and other research tasks.

Harness engineers, do you check every eval run, or only when the pass rate drops?

Try free at @viktor_com. $100 in credits, no card. Full link in my first reply.

Thanks to the team for partnering with me on this post

💬17🔄3❤️37👀8129📊23

Banger paper from NVIDIA. One exciting trend I am seeing is building verifiable environments for yo...

elvis (@omarsar0) — Wed, 07 Oct 2026 15:04:04 GMT

Banger paper from NVIDIA.

One exciting trend I am seeing is building verifiable environments for your agents and training the harness alongside the model.

Co-evolving the harness and the model is a big part of owning the intelligence stack. And many frontier AI companies have started doing that.

This paper shows how this works:

VERA turns benchmark trajectories into more than 9,000 restartable sandboxes with rubric scoring, and keeps only environments that run and can be scored from observable evidence.

It then updates both the model weights and the harness.

A harness edit is kept only if it passes self-tests and adds at least 5 points on the development set.

The system rejects a model checkpoint if its score drops by more than 20%.

At 9B, the co-evolved agent beats the strongest single-axis baseline by 10.3 points on AutoCoWorkBench and 13.0 points on AutoMedBench. At 27B, it scores 71.6 on AutoCoWorkBench, above Claude Opus 4.8.

The environment corpus is open-sourced.

Paper: arxiv.org/abs/2610.05923

Chat with Paper: academy.dair.ai/papers/vera-sc…

💬32🔄44❤️266👀18672📊101

AI voice products need to know who spoke and what they said. That is harder than transcription or ...

elvis (@omarsar0) — Wed, 07 Oct 2026 14:29:09 GMT

AI voice products need to know who spoke and what they said. 

That is harder than transcription or diarization alone, especially when speakers talk over each other.

Pulse by @smallest_AI is now #1 on the Diarization + ASR track of the @voicearena_ai's Diarization Bench, with 24.4% DER.

DER is an error rate, so lower is better. The next system in the track scores 40.7%.

Voice Arena@voicearena_ai

Pulse by @smallest_AI currently ranks #1 in the Diarization + ASR track by DER, as the newest entry on the Voice Arena Diarization Bench.

The Diarization + ASR track ranks Speaker-Attributed ASR systems that return speaker labels together with the transcript. Seven systems are in it today. Pulse scores 24.4% DER at a strict 0 ms collar. The next system, ElevenLabs Scribe v2, scores 40.7%. That is 1.7x lower error than the next best in the track.

Most of the gap is missed speech. For every other system in the track, missed speech is the largest part of the error, above 30% for each of them. Pulse misses 4.4% of reference speech, the lowest in the track.

On the full board of 13 systems, dedicated diarization models and Speaker-Attributed ASR systems together, Pulse places 5th. It is the highest-ranked Speaker-Attributed ASR system on the board and sits ahead of two dedicated diarization models.

The result holds across settings: 
In-person recordings : 24.7% DER
In online calls :  22.4% DER

Congratulations to @smallest_AI 👏

View full results, per track and per condition:voicearena.com/diarization-be…R

Your browser does not support the video tag.

💬11🔄5❤️49👀8278📊16

Introducing MCP tools for @dair_ai. Connect the MCP to your Codex/Claude/Grok Bot to discover and e...

elvis (@omarsar0) — Wed, 07 Oct 2026 13:39:14 GMT

Introducing MCP tools for @dair_ai.

Connect the MCP to your Codex/Claude/Grok Bot to discover and explore the top AI papers. 

We carefully curate the papers, so you can expect mostly bangers. 

The index has all the papers I have featured on X over the last couple of years. 

Get started: academy.dair.ai/papers

> find and summarize awesome papers
> explore our harness engineering collection
> find new methods to experiment with
> find sota results around AI topics
> generate literature review/surveys
> build your personal collection
> visualize papers
> and much more

The demo shows how to install it, including some ideas inspired by @karpathy for visualizing a paper. 

We are just getting started. Reach out if you have feedback. We are releasing a few benchmarks with this in the next couple of weeks and will ship regular additions and improvements. Stay tuned!

Your browser does not support the video tag.

💬26🔄15❤️83👀11885📊44

OpenAI Decisions API just went live, so now is the perfect time to run a more detailed comparison. ...

elvis (@omarsar0) — Tue, 06 Oct 2026 23:16:45 GMT

OpenAI Decisions API just went live, so now is the perfect time to run a more detailed comparison. 

We'll soon find out which LLM-as-a-Judge is better.

I am also comparing it with Perplexity's Decisions API too. So excited to dig deeper into this. More soon.

x.com/OpenAIDevs/sta…

OpenAI Developers@OpenAIDevs

Let your app choose the right model, tool, or action in near real-time with Decisions API, now available to all developers in public beta.

The Decisions API makes decisions up to 10x faster than GPT-6 Luna through the Responses API.

Your browser does not support the video tag.

💬4🔄0❤️4👀1782📊3

Indeed. I would add that if you know how to build a good eval on top of it, you will be on the front...

elvis (@omarsar0) — Tue, 06 Oct 2026 20:13:04 GMT

Indeed. I would add that if you know how to build a good eval on top of it, you will be on the frontier in your domain/task in no time. That is the kind of edge that evals can unlock for you. 

Learn to build good evals too. It's worth it.

Garry Tan@garrytan

You can basically work with agents to write markdown skills and put it on a cron job and do almost every useful type of knowledge work now

💬15🔄5❤️57👀10185📊23

try it here 👇 https://t.co/muf4kpUNpI

elvis (@omarsar0) — Tue, 06 Oct 2026 18:32:35 GMT

try it here 👇agentid.comI

💬1🔄0❤️4👀1315📊2

Great work by @shaped and @parsewaveai.

elvis (@omarsar0) — Tue, 06 Oct 2026 18:21:11 GMT

Great work by @shaped and @parsewaveai.

💬0🔄1❤️7👀1309📊1

We need more efforts like this. Every agent benchmark should audit its verifiers. Parsewave went ...

elvis (@omarsar0) — Tue, 06 Oct 2026 18:12:44 GMT

We need more efforts like this. 

Every agent benchmark should audit its verifiers.

Parsewave went through all 600 public tasks in Zapier's AutomationBench.

Agents wrote realistic wrong answers to try to fool each verifier, and human review confirmed 206 real bugs. AutomationBench Verified fixed all 206.

Regarding 1,235 Kimi K3 runs, the fixed verifiers changed 27.9% of the grades.

I just started looking into this benchmark for some independent eval work I am doing, so this is good timing to see this audit.

shaped@shaped

AutomationBench Verified is out. 

We went through all 600 public tasks in @Zapier's AutomationBench and checked every verifier.

• Agents flagged 323 verifiers as suspicious, human review confirmed 206 real bugs, all 206 are fixed.
• We replayed 1,235 Kimi K3 runs on the old and fixed verifiers and 344 of them (27.9%) got a different grade
• Where verifiers were too strict, pass rate went from 18.8% to 43.8%
• Where they were too lenient, it dropped from 60.2% to 49.7%

Audit and dataset links below.

💬27🔄8❤️47👀10117📊27

Agents need their own identity. This is long overdue. I build a lot of personal agents, and today t...

elvis (@omarsar0) — Tue, 06 Oct 2026 18:02:36 GMT

Agents need their own identity. This is long overdue.

I build a lot of personal agents, and today they either borrow my logins or get blocked at sign-up.

I already use @agentmail, so this is a no-brainer.  

AgentID gives each agent its own verified inbox to sign in with, linked to a real human owner. Apps can let agents in and know exactly who is behind each one.

Haakam Aujla@haakamaujla

Introducing AgentID by @agentmail, "Sign in with Google" for AI Agents

Millions of agents already use AgentMail. AgentID lets them sign in to your app.

Integrate with the prompt below, reply with a screenshot, and we'll give you 3 months free of any AgentMail plan.

Your browser does not support the video tag.

💬8🔄4❤️26👀10114📊13

We need more efforts like this. Every agent benchmark should audit its verifiers. Parsave went th...

elvis (@omarsar0) — Tue, 06 Oct 2026 17:52:22 GMT

We need more efforts like this. 

Every agent benchmark should audit its verifiers.

Parsave went through all 600 public tasks in Zapier's AutomationBench.

Agents wrote realistic wrong answers to try to fool each verifier, and human review confirmed 206 real bugs. AutomationBench Verified fixed all 206.

Regrading 1,235 Kimi K3 runs with the fixed verifiers changed 27.9% of the grades.

I just started looking into this benchmark for some independent eval work I am doing, so this is good timing to see this audit.

shaped@shaped

AutomationBench Verified is out. 

We went through all 600 public tasks in @Zapier's AutomationBench and checked every verifier.

• Agents flagged 323 verifiers as suspicious, human review confirmed 206 real bugs, all 206 are fixed.
• We replayed 1,235 Kimi K3 runs on the old and fixed verifiers and 344 of them (27.9%) got a different grade
• Where verifiers were too strict, pass rate went from 18.8% to 43.8%
• Where they were too lenient, it dropped from 60.2% to 49.7%

Audit and dataset links below.

💬2🔄0❤️6👀1129📊2

This is a great product. Recommended if you are building personal agents and need reliable, fast bro...

elvis (@omarsar0) — Tue, 06 Oct 2026 17:35:16 GMT

This is a great product. Recommended if you are building personal agents and need reliable, fast browser-use tools.

Agent Browsers by @gumloop opens a browser in its own cloud sandbox when there is no connector, then logs in and downloads the export. You do the task once; it saves the steps as a recipe and repeats the job every week after that. 

Great, reliable, reusable browser you can hook all your agents to.

Gumloop@gumloop

Agent Browsers are live on Gumloop today so your agents can reach the work that doesn't have an MCP or API

> Secure credential storage & 1Password integration
> Session replays to see your agents work
> Stealth, proxied sessions

All completely model agnostic

Try Browser today.

Your browser does not support the video tag.

💬2🔄5❤️32👀10006📊9

Own your intelligence stack, folks! Start by tapping into the rich knowledge in your agent traces. ...

elvis (@omarsar0) — Tue, 06 Oct 2026 17:21:52 GMT

Own your intelligence stack, folks!

Start by tapping into the rich knowledge in your agent traces.

Then train small, specialized models to fix production issues. 

@OvermindLab is open source and makes this practical. It builds datasets from your agent's traces, evaluates them against your current model, and trains a smaller specialized model you own.

Overmind@OvermindLab

Overmind turns anyone into an AI lab.

• Builds a context graph from code and traces
• Curates training and eval datasets
• Evaluates prompts and models
• Trains smaller, specialized models

github.com/overmind-core/…

Your browser does not support the video tag.

💬16🔄9❤️130👀70489📊53

Really nice paper on harness-aware distillation for small language model agents. (bookmark it) It'...

elvis (@omarsar0) — Tue, 06 Oct 2026 16:26:02 GMT

Really nice paper on harness-aware distillation for small language model agents.

(bookmark it)

It's a super interesting distillation framework for small language model agents that are deployed with a harness.

In simple terms, you run the big model with and without harness info, then train the small model on cases where its action changes.

Researchers show that adding the harness to on-policy distillation raises how often the student uses harness information on ALFWorld (65.7% to 73.1%) but leaves success flat (43.1% to 43.5%).

Their method, Harness-Aware Distillation, queries the same teacher with and without the harness information and trains the student to prefer the action chosen with it. A filter drops pairs whose preferred action contradicts the harness records. The method uses no task rewards or success labels.

The student reaches 63.4% on unseen ALFWorld tasks, compared with 47.0% for the best baseline, and exceeds its 8B teacher. It also escapes 59.7% of stalls, while the baselines stay near the untrained student's 46.8%.

Paper: arxiv.org/abs/2610.02858

Chat with Paper: academy.dair.ai/papers/harness…

💬16🔄30❤️187👀11485📊68

As I said in the article, one interesting aspect of Jev is the consistent property. That's really us...

elvis (@omarsar0) — Tue, 06 Oct 2026 16:23:32 GMT

As I said in the article, one interesting aspect of Jev is the consistent property. That's really useful for building reliable judges for agentic workflows. I am now working on a little benchmark to test this more broadly and compare it with other decision APIs. More soon.

💬3🔄0❤️7👀2920📊3

I've recently worked on agentic enterprise integrations, and things can get messy fast. You quickly...

elvis (@omarsar0) — Tue, 06 Oct 2026 16:12:59 GMT

I've recently worked on agentic enterprise integrations, and things can get messy fast.

You quickly find that agents are only as useful as what they can do inside a customer's Salesforce or NetSuite.

Every customer sets those up differently, with their own custom objects, fields, and permissions. That means there are many personalized solutions. 

@WithAmpersand is integration infrastructure built for that problem. You define each customer's objects and mappings in one config, and Ampersand handles auth, retries, backfills, and monitoring behind the scenes.

Ayan Barua@ayanb

Your AI is vaporware without deep integrations with your customers’ systems of record.

Introducing @WithAmpersand: integration infrastructure for enterprise agents.

We power 11x, Orb, and Square's ability to take agentic action in systems of record like Salesforce, SAP, NetSuite, and Workday.

Ask anyone serious about building AI and they’ll tell you:

- The SaaSpocalypse didn’t happen. The world depends on CRMs, ERPs, HRISs, and ITSMs.
- Your customers customized their deployments beyond recognition.
- Systems of record companies are basically monopolies. They never had to make their APIs and MCPs user-friendly.
- Docs don't explain half the weird edge cases you'll hit.
- One bad write can blow up your pilot or renewal.
- Once you finally get it working, someone changes a field and it breaks again.

The world's data lives in structured databases. AI needs a translation layer to work with it.

So, for your AI to truly transform the way enterprises work, you'll need deep integrations built for:
- scoped permissions
- bi-directional actions
- real-time speed for agents
- custom objects, fields, and workflows for each of your customers

We built Ampersand to power the future of software.

AI didn't trivialize writing integrations. It made them the critical path.

I believe that deeply, so I didn't make a launch video about Ampersand.

Instead, it's the best builders I know explaining just how big of a challenge this is.

Your browser does not support the video tag.

💬4🔄1❤️23👀8854📊6

@pidotdev Looks like this got a lot of interest. I have added a lot since this version, including co...

elvis (@omarsar0) — Tue, 06 Oct 2026 15:32:37 GMT

@pidotdev Looks like this got a lot of interest. I have added a lot since this version, including connectors to test many proactive capabilities. Also improved memory a bit too. More updates soon.

💬1🔄0❤️5👀1461📊2

Not so impressed with the performance, tbh. I suspect many will feel the same given what we have ...

elvis (@omarsar0) — Tue, 06 Oct 2026 14:55:40 GMT

Not so impressed with the performance, tbh.  

I suspect many will feel the same given what we have recently seen with Kimi, Qwen, DeepSeek, GLM, etc.

But you know what. That's okay!

What we need is more efficient, diverse, customizable frontier open-weight models.

It's great to see Mistral back in the game. 

I like what I see in the multimodal capabilities. That's where I think Mistral Large can stand out. Cyber could also be an interesting direction for them.

Mistral AI@MistralAI

Meet Mistral Large 4, aka Le Chonk. 

• 1T parameters, natively multimodal. 49B active.
It is the best open weights model from US or Europe on aggregated benchmarks.
• State-of-the-art on critical workloads, including cyber defense, manufacturing and finance and it surpasses closed frontier models on visual grounding.
• Forged in Europe end-to-end and is deployable from Europe via our own Mistral Cloud infrastructure.
• Available to all via API today. Working with cybersecurity partners privately.

Open weights release end of October.

Your browser does not support the video tag.

💬24🔄4❤️88👀12279📊24

We built an entire interactive tutorial and a follow-up hands-on lab if you want to dive deep. http...

elvis (@omarsar0) — Tue, 06 Oct 2026 14:12:48 GMT

We built an entire interactive tutorial and a follow-up hands-on lab if you want to dive deep.

academy.dair.ai/resources/jev-…

💬2🔄1❤️5👀2056📊3

This is one of Jev's most impressive use cases. I work a lot on agent evals, judges, and verifiers...

elvis (@omarsar0) — Tue, 06 Oct 2026 14:08:41 GMT

This is one of Jev's most impressive use cases. 

I work a lot on agent evals, judges, and verifiers. 

I've found that Jev-as-a-Judge improves LLM judge reliability through consistency. Makes it ideal for judges, verifiers, and continuous monitoring.

elvis@omarsar0

x.com/i/article/2107…

💬39🔄52❤️541👀59843📊191

https://t.co/j6lRVrWpAZ

elvis (@omarsar0) — Tue, 06 Oct 2026 14:05:07 GMT

x.com/i/article/2107…

💬27🔄64❤️434👀87968📊197

I can usually tell when a deck was made with AI. Generating slides with AI is easy now. They are ju...

elvis (@omarsar0) — Tue, 06 Oct 2026 13:51:22 GMT

I can usually tell when a deck was made with AI.

Generating slides with AI is easy now. They are just missing that personal touch.

Gamma 5 from @GammaApp was rebuilt to address this.

It's really good; the slide decks now feel like my own, with my vision, and closely follow my brand style.

Grant Lee@thisisgrantlee

Everyone is sick of generic AI, including us.

Gamma was the first AI presentation platform to reach real scale. But now that AI tools are everywhere, everything looks the same.

Earlier this summer, we came to an uncomfortable conclusion: without a step change in visual variety, Gamma’s output would look and feel too similar to all the other AI tools. The entire point of a presentation is to help someone "get" your idea, to see it as unique and do something with it.

We started over. Today we're shipping Gamma 5, the biggest product overhaul in our company's history.

We rebuilt Gamma to create presentations that actually look like you and your brand. Your teams can design anything they dream up, following your brand's aesthetic or something entirely new, with every frontier and image model under the hood. And the same goes for your docs, social assets, and graphics.

We revamped our agent, design tools, editing, import, export, connectors, and more.

Get your ideas out there.

Your browser does not support the video tag.

💬2🔄0❤️12👀5462📊3

Like Jev, I believe RL will unlock several more like this, slashing the cost of critical agent opera...

elvis (@omarsar0) — Tue, 06 Oct 2026 00:56:45 GMT

Like Jev, I believe RL will unlock several more like this, slashing the cost of critical agent operations.

flow-1 is competitive in performance, but a huge cost-saver for finding failures in agent traces.  

RSI doesn't only apply to general intelligence. It will equally accelerate specialized intelligence.

Robert@skull8888888888

Introducing flow-1, our new model trained with RL to find errors in agent traces.

It matches GPT-6-sol in trace intelligence while being 23x cheaper. It also costs 25% less to run than GPT-6-luna.

flow-1 finally makes it possible to monitor and understand every agent run, without sampling. 1/6

Your browser does not support the video tag.

💬22🔄18❤️235👀23166📊63

Own your personal agent, folks. Just built my own little personal agent with @pidotdev's Pi Durabl...

elvis (@omarsar0) — Mon, 05 Oct 2026 22:11:43 GMT

Own your personal agent, folks. 

Just built my own little personal agent with @pidotdev's Pi Durable.

I wanted to see how quickly I could build an OpenAI dots clone with open-source tools. 

I am amazed at how quickly I got this off the ground. 

It's proactive. 

It keeps working when you're not watching. It remembers you, runs jobs on a schedule, waits for your approval, and uses its own computer. 

Pi Durable makes that possible.

Every step is checkpointed. I can kill the process mid-task, and My Pi picks up right where it left off.

Memory lives in durable documents. What My Pi learns about me in one Space survives a crash and carries over to every other Space.

Approvals are durable too.

It also has its own computer. Each Space gets a full Linux desktop that My Pi controls with screenshots, mouse, and keyboard. 

The model runs through OpenRouter, so I can swap models without changing anything else. That's all configurable however you like it. 

If you've been wanting your own personal agent, now is a great time to build one. 

I am building this as an experimental ground where I can test new ideas and experiment deeply with proactive personal agents. But I plan to open-source it after testing it a bit and adding more interesting primitives. 

The goal is to make it customizable and easy to build on, so others can experiment with their own personal agent functionalities.

Your browser does not support the video tag.

💬61🔄32❤️384👀33876📊185

Very good to see more competent labs training open-weight models. This Reflection's new Beam model...

elvis (@omarsar0) — Mon, 05 Oct 2026 20:35:47 GMT

Very good to see more competent labs training open-weight models. 

This Reflection's new Beam model seems to have strong reasoning efficiency, which I think is a big deal for long-running agents.

3-4x better inference efficiency than rivals like GLM 5.2, positioning it as a strong Western open-weight option.

Reflection@reflection_ai

Introducing Beam: a highly efficient agentic open model with 501B total parameters and 23B active.

- Frontier reasoning efficiency
- Advances the Western open frontier on coding & agentic tasks
- Trained end-to-end from scratch

Full weights release this month.

Learn more about Beam: reflection.ai/beam

💬12🔄8❤️60👀9092📊16

Recommended. LLM agents love structure, so it's no surprise that a corpus improves agentic search.

elvis (@omarsar0) — Mon, 05 Oct 2026 16:33:55 GMT

Recommended. LLM agents love structure, so it's no surprise that a corpus improves agentic search.

DAIR.AI@dair_ai

Banger paper from Microsoft and colleagues.

If you run agents that search a large document collection, this one is worth your time.

(bookmark it)

They introduce CorpusMap, which resolves recurring entities across the collection in advance and gives each entity a page that links to every document that mentions it.

The original documents stay in place.

The agent reads a document, follows an entity to related documents, and avoids searching for the same evidence again.

Across 7 models and three benchmarks, answer quality goes up 6.4 to 11.7 points while input tokens drop 34% to 57%.

It also beats an LLM Wiki layer and three other navigation layers.

The map can be built without LLM calls and updated as new documents arrive.

Paper: arxiv.org/abs/2609.37226

Chat with Paper: academy.dair.ai/papers/follow-…

💬19🔄15❤️216👀33191📊72

Pay attention to this if you are building with agents. Simulated companies are a big deal for impr...

elvis (@omarsar0) — Mon, 05 Oct 2026 16:22:20 GMT

Pay attention to this if you are building with agents. 

Simulated companies are a big deal for improving agentic products.

This is because good agent evals need environments that behave like real companies.

Era by Eon can generate a complete company across Salesforce, Zendesk, Slack, Jira, and other systems.

Each system behaves like the real vendor, down to rate limits, pagination, and error codes. You can reset the company and rerun the same tasks after changing the model or the prompt.

Ofir Ehrlich@OfirEhrlich

You can’t safely test an enterprise agent on a real company’s data.

So we built a company for it to work in.

Era is live today, and it’s free.

It generates a complete simulated enterprise that behaves like a real one across Salesforce, Slack, Jira, Zendesk, Gong, Deel and more, along with cloud databases and storage. Agents interact with it through live MCP and API interfaces.

People leave. Deals change. Records get duplicated. Permissions differ across systems. And because Era generated the company, it knows the exact ground truth.
Test, benchmark and improve agents against realistic enterprise workloads, use the failures for targeted post-training, then rerun the same environment to measure the impact.

Huge thanks to our research partners @NVIDIA, @Decart, @Composio, @openlayerco, @Deel, @Eragon and @Plurai, with more coming soon.

Your browser does not support the video tag.

💬8🔄4❤️38👀9870📊12

A big milestone for AI in healthcare. @nollahealth can now use AI to prescribe acne treatment in Ut...

elvis (@omarsar0) — Mon, 05 Oct 2026 15:24:47 GMT

A big milestone for AI in healthcare.

@nollahealth can now use AI to prescribe acne treatment in Utah. The AI handles intake, a skin scan, assessment, prescription, and follow-up, and a clinician steps in when needed.

Starting with acne makes sense to me because it is common and relatively low risk.

Very excited about AI-powered personalized healthcare. I have personally benefited from AI for health, so it's awesome to see this progress.

Luis Wenus@luiswenus

Today, Nolla Health became the first organization in the U.S. (and possibly the world) to receive regulatory approval for an AI to issue initial prescriptions.

This makes Nolla the first ever actual end-to-end AI doctor.

Your browser does not support the video tag.

💬9🔄1❤️50👀9751📊12

Or just build your own. That's how you make personal agents fully yours. You can sit and wait fo...

elvis (@omarsar0) — Mon, 05 Oct 2026 15:08:22 GMT

Or just build your own. 

That's how you make personal agents fully yours. 

You can sit and wait for the next bug fix or for one of these providers to address your safety concerns, or you can start co-evolving a personal agent as needs arise. 

I have chosen the latter. 

A personal agent is something you should fully own, from the traces to memory to all outputs. 

So much resistance to building your own harness and agents. I understand the fear of maintaining yet another thing.

But AI makes it easy to build and maintain this stuff. 

It's not that hard to build agents today, folks. 

You seriously should consider building your own solution. This is a great time to start. Start from scratch, keep it minimal, and evolve it as you go. Or you can start with an open-source solution like Pi Durable and build from there (btw, Pi Durable is crazy good for this stuff).

Kun Chen@kunchenguid

there’s something quite awkward about all the personal agents in the current hype cycle

muse, grok bot, instinct, dots, and whatever google, anthropic will come up with

none of them is “mine”

i’d be trusting a vendor for some of the most sensitive data about me and bet on them handling it with extreme responsibility. what if they have a data leakage incident? what if they hired a wrong employee? what if they simply break bad?

i’d also be betting on their model. right now i’ve set up a lot of my stuff in grok bot, but what if their model falls behind? what if another model becomes 10x better on pure intelligence? then i’ll either miss out on a better assistant, or have to bite the bullet and do a full migration

i’d also have to bet on their product. they may not build the features i want. they may not allow me to customize enough. and one day my assistant may show me ads

that feels like way too much betting than what i’d be comfortable with, if i truly want literally everything in my life to go through and be taken care of by an assistant

with all that considered, i’m more bullish on open source personal agents that run on people’s own computers. but i think these open source projects have to break out of the assumption that their users are developers and they can just throw a repo at them and ask them to launch a terminal

a well polished, fool proof, community maintained, vendor agnostic, free personal agent that everyone can run by themselves without locking into a SaaS - that’s what i think many people will need

💬14🔄3❤️40👀9962📊17

Learn to optimize your own harness, folks. You can squeeze much more performance and value from a h...

elvis (@omarsar0) — Mon, 05 Oct 2026 15:01:16 GMT

Learn to optimize your own harness, folks.

You can squeeze much more performance and value from a harness.

This work claims that a self-modified harness matched Codex for $4.03.

Specifically, a coding agent rewrote its own harness until it solved 82.0% of Terminal-Bench 2.1 with DeepSeek V4 Flash, without any task reward during the search.

The search cost was $4.03.

That matches Codex, the top harness in a public nine-harness comparison run under the same settings.

SelfSearch has agents modify their own instructions, tools, and procedures using records of earlier self-modification attempts. Each record holds the reasoning, tool actions, and outcomes. The modified agent then becomes the next improver.

Population-mean success rises in all six model-benchmark settings, with single agents gaining up to 11.2 points. On SWE-bench Multilingual, one evolved agent gains 5.0 points and spends 38.5% less on tasks both versions solve.

Paper: arxiv.org/abs/2609.37968

Chat with Paper: academy.dair.ai/papers/selfsea…

💬60🔄34❤️261👀19123📊135

These days, I spend more time verifying agent-written code than writing it. This is a big bottlenec...

elvis (@omarsar0) — Mon, 05 Oct 2026 14:02:49 GMT

These days, I spend more time verifying agent-written code than writing it.

This is a big bottleneck in so-called software factories. 

Ship from @ContextQa takes a bug reported in Slack or Linear, reproduces it, and hands Claude Code or Codex the context to fix it.

Coding agents need this kind of feedback loop to get better at fixing their own bugs.

Deep Barot@deepcabinwala

Everyone got a coding agent. Nobody got a QA engineer.

Until today.

Meet Ship, your Autonomous Quality Engineer.

It tests deployed PRs, reproduces bugs from Slack and Linear, and hands Claude or Codex the context to fix them.

Try it free: ship.contextqa.com

Your browser does not support the video tag.

💬22🔄2❤️24👀9562📊18

Great overview of context compression in LLM Agents And one interesting, unexpected finding. If yo...

elvis (@omarsar0) — Mon, 05 Oct 2026 02:00:04 GMT

Great overview of context compression in LLM Agents

And one interesting, unexpected finding.

If your compaction policy is tuned to cut tokens, it may be making your agent run slower.

Great study from UT Austin on context compression in coding agents.

They ran nearly 35,000 agent runs on SWE-bench Verified and Terminal-Bench and varied three decisions separately. These are how context is compressed, when compression triggers, and how much is removed.

On Terminal-Bench with Qwen, policies that use about a third of the tokens can take 20% to 80% longer than keeping full context.

Step-triggered policies cut the most tokens per step but need 10% to 27% more model calls. Threshold-triggered policies cut tokens by 22% to 55% with call counts close to full context.

Results also differ by model. A policy that works well for Qwen drops Devstral to 38.7% and makes it slower, so measure latency and cost per model before choosing one.

Paper: arxiv.org/abs/2609.32961

Chat with Paper: academy.dair.ai/papers/beyond-…

💬63🔄31❤️171👀14711📊97

50K likes on this!? Most of what Karpthy has shared is things we have been doing for over a year. ...

elvis (@omarsar0) — Sun, 04 Oct 2026 17:40:39 GMT

50K likes on this!?

Most of what Karpthy has shared is things we have been doing for over a year. 

No surprises, as many replies are highlighting.

But I still think this raises an interesting discussion. 

Is there a universal interface/langauge that allows more seamless communication between agents and humans?

I believe there is. And I think that is what @karpathy is suggesting we think about. 

Right now, ideas are all scattered. 

As some of you know, I have been early on LLM Wikis, Artifacts, AI-generated visual explainers, dynamic interactives, etc. 

Over the past couple of months, I have been experimenting with a new universal interface for seamless human-agent collaboration. So I have a lot to share, but it's still too early, and I frankly haven't found the perfect solution. 

In short, my current optimal setup (see a simple snapshot attached) ties my high-level agent to highly flexible Notion-like pages that support embedding artifacts, text, videos, visual explainers, and just about anything an LLM can generate. And if it doesn't support it yet, I ask it to build it on the fly. It's crazy sometimes, the features it builds, but this is a co-evolved interface that satisfies both my agent's and my needs. 

The interface supports code review, note-taking, writing, research, prototyping, mock designing, and so on.  

In fact, I have now wired my agents to intelligently spit out these artifacts (for lack of a better word), depending on the task/instruction, instead of the default text responses, which are frankly overwhelming and hard to read at times. In other words, as I keep building this "universal interface," it feels less and less like I am chatting with an agent; I'm collaborating with it directly through the artifacts (through comments and suggestions). It allows me to move faster. I can consume and review faster. I am better organized. It feels like a language both the agent and I can speak fluently, so there is very little friction. 

Listen, this is all a personal experience. I don't think everyone will use interfaces like this. But it's been interesting to see Artifacts gain wide adoption, and now dots have Spaces that look a little like what I built here several months ago. 

More to share soon.

Andrej Karpathy@karpathy

We'll be spending a lot more time trying to understand the outputs of language models. A few thoughts, tips & tricks:

Writing. Something I've had success with: Ask your LLM to explain something in ASD-STE100, it's a controlled language specification originally developed for aerospace maintenance documentation. LLMs well-versed in this language and it comes with heavy constraints on clean writing style that I often find a lot more readable. Sometimes I've tried to soften it a bit e.g. ask for "80% of the way to ASD-STE100" because the spec is quite stringent. But even better:

Diagrams / images. Instead of writing, ask your LLM to create a diagram. These can be a lot easier to process, parse, and understand. But even better:

Web pages. Ask for output "in HTML" to get a beautiful, interactive webpage. LLMs are getting really good at frontend and can create beautiful experiences, animations, etc. But even better:

Explainer videos. The output format I am most bullish on is fully custom / bespoke explainer videos generated on any arbitrary topic. Experiment with things like "Create a 3b1b style video explainer on X. Use my ElevenLabs API key for audio narration". (you'd need an API key for the latter or you can ask your LLM to find you decent free alternatives that use your local compute). This is actually starting to work!

In summary:
- As LLMs get better, they will do more and more of the legwork autonomously, and a lot more of our work will rise up the abstractions into oversight and understanding.
- Luckily, LLMs can help here too because as intelligence and code are increasingly abundant, you can ask for large, custom, discardable software artifacts (e.g. web apps, video explainers) that would have never made sense to create before. Push the boundaries here and you'll be surprised.

💬80🔄38❤️729👀200186📊227

Same here. I used the terminal for a long time, and I loved it. Agents have changed all of that....

elvis (@omarsar0) — Sun, 04 Oct 2026 14:23:39 GMT

Same here. 

I used the terminal for a long time, and I loved it. 

Agents have changed all of that. 

I didn't imagine that that day would arrive, but I have completely moved away from the terminal.

You can delegate so much more of this stuff to agents now.

And today there are undeniably better interfaces for working with agents. 

But that experience with terminals has made me a better systems thinker, which I have used in areas like designing useful sandboxes and interesting abstractions on top of them.

Yuchen Jin@Yuchenj_UW

I was a terminal person for 15+ years.

I loved Vim, knew all the shortcuts, and a black terminal made me feel like I was a cool hacker in The Matrix.

But terminals assume humans operate computers directly via files, commands, processes.

Agents changed all that.

You just say your intent. The agent operates the machine, using the same programming languages and Linux commands we spent years learning.

I’m still glad I learned computer systems before the AI era though. Understanding what sits beneath the abstraction makes you a better systems thinker.

That’s a moat in the AI era.

💬24🔄2❤️66👀11829📊25

Tbc, given how much confusion in replies, coding agents still operate via CLI. For me, that's happen...

elvis (@omarsar0) — Sun, 04 Oct 2026 14:12:49 GMT

Tbc, given how much confusion in replies, coding agents still operate via CLI. For me, that's happening in the background now, rather than being the main way I directly interact with agents. I agree with Yuchen in that the terminal era is over. So many constraints with it. Codex Desktop is as good as it gets today, but even that seems not optimal in some cases. So what I have built and use is an interface that's more flexible and lets me combine focused agent sessions with high-level persistent agents that can manage all my agent sessions when the number of them gets out of control.

💬1🔄0❤️4👀2036📊2

Banger paper from Google. It's standard practice to sample several agent rollouts and trust the ans...

elvis (@omarsar0) — Sun, 04 Oct 2026 11:00:11 GMT

Banger paper from Google.

It's standard practice to sample several agent rollouts and trust the answers they agree on.

This Google paper shows that agreement can hide shared errors, while disagreement often points to the correct alternative.

VeriHarness turns the same base model into an agentic verifier with two jobs.

One resolves claims where rollouts disagree by checking workspace evidence.

The other challenges claims that every rollout agrees on and looks for requirements they all missed.

Across five long-horizon benchmarks, it gives the best selection scores among the baselines tested. With evidence-backed revision, it adds 6.2 points over a single rollout with Gemini 3.5 Flash and 6.4 points with Claude Opus 4.8.

The authors also release about 26,000 rollouts.

Paper: arxiv.org/abs/2610.00972

Chat with Paper: academy.dair.ai/papers/verihar…

💬56🔄69❤️447👀30945📊175

New paper from Microsoft and colleagues on evolving agent harnesses. It's a really cool idea to evo...

elvis (@omarsar0) — Sun, 04 Oct 2026 05:37:02 GMT

New paper from Microsoft and colleagues on evolving agent harnesses.

It's a really cool idea to evolve a harness from published research. Something I have also been testing for the past couple of months.

ScholarEvolve proposes harness changes based on published agent research rather than the agent's own failure logs.

It splits the harness into modules for tool use, memory management, and task execution.

It runs topic modeling over recent papers to identify distinct improvement strategies for each module, then implements and tests combinations.

New papers can be added over time.

With the model held fixed, Qwen3.5-27B goal completion on AppWorld Challenge rises from 49.6% to 63.6%, and GPT-5.4-mini on Tau2-Bench Telecom rises from 72.7% to 81.9%.

Paper: arxiv.org/abs/2609.40169

Chat with Paper: academy.dair.ai/papers/learnin…

💬29🔄34❤️191👀13240📊77

Use of coding agents via CLI has been dead for a while now. We are early on discovering better inte...

elvis (@omarsar0) — Sun, 04 Oct 2026 03:58:53 GMT

Use of coding agents via CLI has been dead for a while now.

We are early on discovering better interfaces but interactions now happen directly with agents and will continue that way going forward.

What is working for me is a mixture of high-level persistent agents (like dots) and specialised agents (like codex desktop sessions) that are themselves managed by the persistent agents or accessible at anytime when I need to do focused work.

This provides me lots of flexibility and I can leverage a variety of experiences and interactions with agents that range from the simpleness of grok bot to the more involved sessions offered by codex desktop.

These again are some of the advantages
I enjoy because I have built my own agent orchestrator and harness.

Yuchen Jin@Yuchenj_UW

I haven’t touched Claude Code or Codex CLI in a while.

The terminal era is over imo. It's the wrong interface for coding agents. Tabs are ephemeral, but context is persistent, and managing 30 tabs is pure cognitive overhead.

I don’t really need an IDE like Cursor either. I rarely navigate the whole codebase anymore.

The new primitive is the agent, not the file.

(Codex desktop app is the best agentic UI for now. But we’re still early.)

💬55🔄5❤️171👀33916📊70

Banger paper from CMU on harness learning. (bookmark it) Also, pay attention to this important new...

elvis (@omarsar0) — Sat, 03 Oct 2026 23:38:01 GMT

Banger paper from CMU on harness learning.

(bookmark it)

Also, pay attention to this important new AI engineering skill of improving agents by editing their harness code instead of their weights.

Seeing a huge shift towards this.

The authors train a proposer model with RL to read a task, the current harness and an execution report, then write a code edit to the harness.

The reward is the score of the revised harness. The solver model never changes.

A trained 4B proposer beats its 35B teacher at single-step revision on Reasoning Gym, including task families it never saw in training. A proposer trained on HotpotQA keeps improving harnesses on MuSiQue and 2WikiMultihopQA.

Paper: arxiv.org/abs/2609.35738

Chat with Paper: academy.dair.ai/papers/harness…

💬38🔄86❤️469👀28847📊183

Dear Anthropic, please don't nerf Opus 5.5. I have been impressed by Opus 5.5 (using it via API). S...

elvis (@omarsar0) — Sat, 03 Oct 2026 19:54:32 GMT

Dear Anthropic, please don't nerf Opus 5.5.

I have been impressed by Opus 5.5 (using it via API). So I resubscribed to Max and was just mind-blown by how durable it has become. 

If you used Fable models, you would know how terrible and useless that Max subscription had become. 

Opus 5.5 is obviously one of the best models as a daily driver, but I didn't expect to get so much from the Max subscription with this new model. 

Say what you want about Anthropic, but Opus 5.5 is extremely efficient and a really good model.

It's exciting. I hope models get more efficient and we can squeeze more value out of every token.

💬38🔄5❤️234👀20843📊54

If you want more related reading on this. The mentioned context language model paper: https://t.co...

elvis (@omarsar0) — Sat, 03 Oct 2026 17:17:10 GMT

If you want more related reading on this. 

The mentioned context language model paper: academy.dair.ai/papers/context…

The AutoHarness paper: academy.dair.ai/papers/autohar…

💬3🔄1❤️9👀2184📊5

Great paper for all you pre-training and post-training nerds. The surprising finding is that pre-tra...

elvis (@omarsar0) — Sat, 03 Oct 2026 16:36:33 GMT

Great paper for all you pre-training and post-training nerds. The surprising finding is that pre-trained LLMs, equipped with a light inference harness, can serve as capable agents and can even surpass post-trained counterparts with a sufficient test-time budget.

DAIR.AI@dair_ai

Banger paper from Meta Superintelligence Labs.

They find something super interesting and unexpected.

(bookmark it)

Base models with a light harness often solve more agentic tasks than their RL post-trained versions when both get enough samples.

Post-trained models win on pass@1. At large K, base models frequently solve tasks the post-trained ones never solve on BFCL v4 multi-turn, ACEBench, and WebShop.

This is because post-training pushes each task toward always solved or never solved. Consistency goes up, and coverage goes down.

The authors call the lost test-time scalability the Sharpening Tax. Across 42 base and post-trained pairs, it shows up in most settings, grows with model size, and can be estimated from a few rollouts.

Their fix, PTGS, sets the sampling temperature per prompt from its estimated difficulty during RL. It pays a smaller tax and also raises pass@1.

Paper: academy.dair.ai/papers/sharpen…

💬17🔄10❤️133👀11498📊33

First AutoHarness, then AutoContext, now AutoCompact. I am seeing a rising trend of work that train...

elvis (@omarsar0) — Sat, 03 Oct 2026 16:11:02 GMT

First AutoHarness, then AutoContext, now AutoCompact.

I am seeing a rising trend of work that trains models to natively support more of what the harness does.

This work specifically trains agents to decide for themselves when to compact.

Reminds me of the new paper from Meta that trains models to manage context natively.

But how good is this approach?

AutoCompact trains the agent to decide when to compact, what working state to keep, and how to resume.

A judge first reviews the base agent's compaction decisions and replaces flawed ones before they execute. The corrected trajectories are used for SFT, then RL with task-success rewards trains coding and compaction together.

Pass rates improve by 9.2 points on SWE-bench Verified and 5.0 points on SWE-PolyBench Verified.

The gain holds even with a 256K window that never overflows, so learned compaction helps when context space is not the limit.

It remains to be seen how this works at scale and how robust it is across harnesses.

One interesting note from the authors is that this type of proactive compaction is a form of model-harness co-design: the harness provides the compaction mechanism, while the model learns when to invoke it, what to preserve, and how to continue afterward.

Even more interesting is how to combine the rule-based compaction techniques already packaged in harnesses with more model-invoked proactive ones.

Paper: arxiv.org/abs/2610.02163

Chat with Paper: academy.dair.ai/papers/autocom…

💬44🔄39❤️244👀19726📊103

This is impressive! I cloned my voice in the @cartesia playground from a ~15-second clip. The clone...

elvis (@omarsar0) — Sat, 03 Oct 2026 14:33:32 GMT

This is impressive!

I cloned my voice in the @cartesia playground from a ~15-second clip. The clone was ready in a couple of seconds. 

Then I had it speak Japanese, a language I don't speak. 

My clone now says in Japanese that I can find good papers by combining AI with my experience reviewing papers. 

I am impressed by how good this sounds. 

This runs on Sonic 3.6, their latest voice model, which supports 44 languages. You can do the same with any of them.

This tech is getting really good, and I think it's worth exploring to make your content more accessible to more people in different languages. 

Try it here: play.cartesia.ai
More info about their model here: cartesia.ai/blog/sonic-3.6

Your browser does not support the video tag.