validated_ron

@BVeiseh Your examples are plain agentic conversation though. Better example for agentic loop is agent waiting on an alert queue and remediating as they come

Cybersecurity is fundamentally changing underneath us. And there is a new key method that security teams are using to stay ahead: loops. It is more critical now than ever to start learning how to use the new tools of the trade. Using AI for security is no longer copy pasting alerts into ChatGPT with an MCP connected. In order to stay ahead, learn how to use agents in loops. Loops are incredibly powerful, we use them daily to help us build MindFort, and to empower our security agents to help our customers find vulnerabilities. But what is a loop? A loop is when an agent receives a clear goal, then acts, sees the result, and decides its own next move, repeating until the goal is met. They can be incredibly powerful for security folks as well. Here are some quick examples of how agentic loops can be helpful in security work: - Recon: Point an agent at your external surface and let it enumerate, probe, and re-prioritize as it learns, instead of running a fixed playbook. - Validation: Hand an agent a pile of scanner findings and ask it to actually prove each one, quickly helping you weed out false positives. - Research: Give an agent a set of constraints and methodologies and let it get to work iterating constantly on a new technique or exploit until it figures it out. Loops are powerful, but they're also extremely expensive. To give any of these kinds of loops a try, I'd recommend the following setup: For the model, use GLM 5.2 hosted via Fireworks AI in the US. For the agent, use OpenCode. An excellent open source terminal agent with better performance than stock Claude Code.

@iamlukethedev I feel crafting a good skill/context is where you as an engineer squeezes that extra value from AI. Teaching an AI through an AI generated skill is slop in slop out IMHO

Hermes learned my entire repo in minutes. I’m not exaggerating. I pointed the new /learn skill at my codebase… and watched it turn the repo into a reusable skill. Files. Patterns. Architecture. Commands. Workflows. It didn’t just “read” the repo. It created a playbook for how to work with it again later. This blew my mind. Memory remembers facts. Skills remember how to do the work. That’s the difference. x.com/iamlukethedev/…

@sflorimm Because people got used to AI being imprecise software and adjusted their expectations

nobody talks about AI hallucinating anymore. why?

@4lexsvv Aside from securing the agent to financial data interface where else does it add value over letting Claude read the bank statement?

Yes, AI has made founders dramatically faster. I'm not convinced it's made them better informed. Those two things can drift apart very quickly. That's the gap we built Nume for.

@EHuanglu Question is can it stay consistent between subsequent 30s clips so you can stitch them to a full length? Full video is nice but I think real near term breakthrough will be with limited scope specific asset generation

i dont think people realize how big this is Seedance 2.5 now can generate 30s 4K videos from one prompt.. with up to 50 ref.. one click filmmaking is here

@thaiscbranco_ Is that like synthetic data or real world captured?

i know someone that makes good data

@TheSeanRich @LangChain Will be awesome if at some point will be able to integrate Claude Code as a "deep agent" into LangGraph. Got serious mileage and trust with Claude. Feels the lack of structured workflow and observability like LangGraph's though

if you want to be on the forefront of agent engineering, you have to pay attention to deep agents and the research coming out of @LangChain

@itsnicholash It's a leap of faith giving the agent expanded autonomy. Probably easier with good observability

Fable can run autonomously for 4 hours But what about 4 days? Parvind from Redscope AI teaches you exactly how to build long running agents And WHY it’s valuable. Bookmark it

@Stillm4n @ABNAMRO @Deloitte @donginsel Sounds interesting. Probably need serious guardrails around those agents if you want to see your money back though

Kicking off Dutch Blockchain Week strong with @ABNAMRO‘s Anaelle Ubaldino, @Deloitte‘s Kim Schneider and @donginsel at Agentic Day tomorrow. If you are curious about the convergence of stablecoins and AI this will be the panel for you:

Unit testing is the executable extensions of manual code reviews. As AI pushes developers further away from hands-on contact with code the testing methodology need to catch up

Relying solely on recorded real world datasets for agentic AI evaluation means reactively chasing incidents i.e. testing in production

If Claude Mythos/Fable can build an entire web app as rapidly as a one-shot prompt how does testing catches up so developer still feels in control of this unchecked explosive growth

@heyDhavall Forced to only look at the system E2E how does the dev lays out the required edge cases?

Code generation is no longer the bottleneck. Validation is. AI can generate hundreds of lines of production code in minutes. The problem is that PR reviews haven't scaled at the same pace. Most teams are seeing the same thing: → More code shipped per sprint → More AI-generated PRs → More pressure on reviewers → More bugs slipping through AWS CTO Werner Vogels recently called this verification debt. That's why @Test_Sprite caught my attention. Instead of reading a diff and guessing what might be wrong, it opens the actual application and uses it. It explores user flows, generates test plans, executes tests in parallel, and surfaces issues with: → Error → Trace → Cause → Fix What stood out to me is the new portal experience. You can actually watch agents explore the product, inspect failures, replay sessions, and trace API behavior end-to-end. For engineering teams, that's a much more useful signal than another AI telling you what it thinks about your code. Code review still matters. But it shouldn't be the only line of defense between AI-generated code and production. 🔗 testsprite.com

@eric0xbt Since current toolset already support this limiting factor is enclosing harness/workflow reliability which can take years to iterate on until is right and businesses start to trust it. Maybe more like 2035+

Twelve months from now, at least 3 AI agents will generate more revenue than entire teams of 50+ people in digital marketing agencies. I didn't believe this until recently. Then I started seeing early examples everywhere, and by June 2027, it will be a baseline reality. I’m putting my name on this prediction publicly. @RallyOnChain What convinced me was seeing one person with the right AI stack accomplish work that previously required multiple specialists. That trend isn't slowing down. The advantage isn't intelligence alone. It's that AI agents can operate continuously, scale instantly, and cost a fraction of a traditional team. Companies that adopt autonomous AI workflows early will operate with dramatically smaller teams and higher margins. Bookmark this and hold me accountable in June 2027.

@pk_iv Dry run simulation testing

every agent platform needs to offer: 1. sandbox 2. model router 3. observability 4. ??? what am i missing?

@coreyganim Don’t try to teach an AI how to fill the role of a human. Lean on it’s unique differentiating abilities and find new ways to get the job done.

Do not sell "time saving" AI solutions. Everyone is doing this and the competition is brutal. Instead, sell AI solutions that drive revenue. - Speed-to-lead - Quote automation - Missed-call text back - Sales follow-up workflows - Lead qualification systems These solutions make the business more money, which means the ROI is virtually limitless. When you sell time saving solutions, the ROI is capped because there is only so much time a business owner can save in a week. But they can make an infinite amount more money. Keep this in mind next time you go to pitch an AI solution.

@atulkumarzz When AI is testing AI I think the question becomes how do you give tools to keep the developer in the loop enough so he can maintain ownership

Everyone's talking about AI coding. Almost nobody is talking about what happens after the code gets generated. That's where things get messy. AI can crank out features in minutes, but testing, validation, and QA are still operating at human speed. The result? More code shipped, more bugs escaping, and more late-night fire drills. That's why @Test_Sprite stood out to me. Instead of just reviewing code, it actually interacts with your application like a real user. It clicks through flows, explores edge cases, generates test plans automatically, and catches issues before customers do. The new web portal takes it even further: → Parallel AI agents exploring different user journeys → Visual test replays so you can see exactly what happened → Auto-Heal that adapts when the UI changes → End-to-end API and data-flow visibility The shift isn't AI writing code. The shift is AI becoming your first QA engineer. And that feels like a much bigger deal than most people realize.

@hayyantechtalks Trying to understand where this tool adds value compared to the infinite flexibility of manually asking ChatGPT to generate these items?

I came across a really interesting video today. A solo founder building an entire brand in under 30 minutes: ✓ Logo ✓ Landing page ✓ Social content ✓ Marketing assets ❌ No designer ❌ No agency ❌No creative team Just one AI tool and a few prompts. I had to test it myself. Here's what happened 👇

@ypatil125 Think this is kind of romanticising. Worked with FDE in other domains where they are common. Usually not the best and brightest then they badly integrate with the team and the added value is pretty slim

The real power of forward deployed engineering has always been putting strong technical people directly alongside the operators who own the outcome. That proximity forces the work to solve the actual problem instead of some sanitized version of it. In the AI era this principle has become even more valuable. Agents can now sit inside real workflows and improve from actual decisions, which means the highest-leverage work is extracting the tacit knowledge that lives with subject matter experts, building evaluations that reflect how things actually break, and closing the production feedback loop so agents get better from real outcomes.

@unclebobmartin Maybe just go along with their identity crisis. Who is saying an AI agentic org needs the same roles as a human org does

I've been allowing the swarm to operate for a full day now. It's gotten a lot done. But I've intentionally allowed the agents to continue compacting their contexts, over and over. The results are interesting. First, they've started to lose their particular identities. The coders is doing crap analysis. The refactorer finds it has nothing to do because the coder did it. The refactorer has decided that sending message to other agents requires my permission (it doesn't). The coder simply forgets to hand off to the refactorer until I remind it. The architect decided to not run mutation tests because they take a long time. Fascinating. Given all that, and the constant babysitting that is required, it has gotten a LOT done. As the number of testing scenarios has increased, the testing procedures have gotten quite slow. The continual retesting is very inefficient. I'm going to have to implement impact analysis to drive the tests so that only the things that have changed are tested.