Kyriakos

Kyriakos

Users
Tweets

Kyriakos

@Kyriakos_Pelek

Replying to @ipfconline1 @InformationWeek @DeepLearn007 @Nicochan33 @Ym78200 @AkwyZ @kalydeoo

Solid advice, focus on real need

Alejandro ♫⋆

Alejandro ♫⋆@magodelosdatos

Replying to @ipfconline1 @InformationWeek @DeepLearn007 @Nicochan33 @Ym78200 @AkwyZ @kalydeoo

Vertical specialization proprietary data. That's the moat. Same in autonomous systems: own the pipeline, not just the model. The market is finally valuing substance over hype.

J.P. Challenger

J.P. Challenger @PowerInVOX

Replying to @ipfconline1 @InformationWeek @DeepLearn007 @Nicochan33 @Ym78200 @AkwyZ @kalydeoo

En 2026, l'IA crée de la valeur lorsqu'elle part d'un besoin réel, avec une vision claire, une exécution rigoureuse et une confiance durable. C'est ainsi que l'innovation devient une vraie transformation.

Saku013

Saku013 @saku013

Replying to @RapidReport2025 @DeepLearn007

But why even do this at all. What a waste of life

ipfconline

ipfconline @ipfconline1

It's the year of the #AI app: Tips to build a successful one buff.ly/cqyIYk3 @InformationWeek #ResponsibleAI Cc @DeepLearn007 @Nicochan33 @Ym78200 @akwyz @kalydeoo

171

Terence Leung

Terence Leung

@TerenceLeungSF

18h

Appreciating this world: Witnessing a #bff full of #excitement and #happiness We are #happier as a result too… @mvollmer1 @morgfair @ChuckDBrooks @Nicochan33 @enricomolinari @NancySinatra @Ronald_vanLoon @alvinfoo @KirkDBorne @Hana_ElSayyed @JimHarris @MikeQuindazzi @Shi4Tech @mhcommunicate @ipfconline1 @kashthefuturist @rwang0 @HeinzVHoenen @YuHelenYu @BetaMoroney @antgrasso @kuriharan @PawlowskiMario @EvanKirstel @HaroldSinnott @terence_mills @FrRonconi @TamaraMcCleary @UrsBolt @pascal_bornet @HeinzVHoenen @SpirosMargaris @richardturrin @Xbond49 @psb_dc @rshevlin @JimMarous @IanLJones98 @Khulood_Almani @enilev @GlenGilmore @DeepLearn007 @KamLardi @debashis_dutta @sallyeaves @EstelaMandela @NevilleGaunt @IngridVasiliu @Eli_Krumova @baski_LA

0:21

786

Dan Mair

Dan Mair @maircomm

22h

Replying to @AlistairCarns @DeepLearn007

It's not stopping AI from writing your posts though.

Terence Leung

Terence Leung

@TerenceLeungSF

Jun 13

Most teams using #AI agents are still operating in a mode that keeps humans squarely in the driver’s seat. Every cycle requires someone to trigger the next action and then review the results before anything else can happen. This manual oversight quickly becomes the real constraint on scale and consistency. Andrej Karpathy highlighted a more powerful direction when he noted that the real opportunity lies in removing yourself as the limiting factor: so a small amount of input can trigger a large amount of autonomous progress. Loop Engineering is a practical way to make that shift. Instead of relying on constant human intervention, you build a structured system that lets agents handle both the execution and the quality control on their own. The architecture centers on four key elements working together: • A scheduler that determines the next piece of work to tackle • A core execution agent that carries out the tasks • A dedicated reviewer agent that evaluates results independently and supplies focused guidance for improvement • A shared, persistent record stored outside the model’s context (such as a structured file or graph) that both the executor and reviewer can access and update across multiple runs This setup allows the process to continue without interruption until clear stopping criteria are reached: whether that’s successful completion, a maximum number of cycles, or a resource limit. To make these systems dependable in practice, several design choices matter: • Keep the reviewer role separate so it can catch issues rather than simply confirming what was already done. • Establish clear stopping rules before any run begins, rather than trying to decide mid-process. • Store progress and context externally so the system can pick up exactly where it left off, even after long pauses. • Begin with straightforward, easy-to-verify checks before moving to more complex judgment calls. Of course, giving agents more independence also introduces new challenges around overconfidence in results and gradual drift in the surrounding instructions and tools. The strongest implementations address this by adding automated feedback loops that turn production observations into diagnosed improvements, verified corrections, and lasting safeguards against repeat issues. The end result is agent workflows that deliver compounding value with far less day-to-day oversight. I’ve included a clear diagram below that shows the contrast between traditional hand-managed sessions and this more autonomous approach. What parts of your current AI agent work still require the most ongoing human attention? Curious how others are reducing that dependency in their own setups. Please also give feedback on how to improve the diagram below! #AIAgents #AgenticAI #AI #ArtificialIntelligence #SoftwareEngineering #Productivity #FutureOfWork #MachineLearning @mvollmer1 @morgfair @ChuckDBrooks @Nicochan33 @enricomolinari @NancySinatra @Ronald_vanLoon @alvinfoo @KirkDBorne @Hana_ElSayyed @JimHarris @MikeQuindazzi @Shi4Tech @mhcommunicate @ipfconline1 @kashthefuturist @rwang0 @HeinzVHoenen @YuHelenYu @BetaMoroney @antgrasso @kuriharan @PawlowskiMario @EvanKirstel @HaroldSinnott @terence_mills @FrRonconi @TamaraMcCleary @UrsBolt @pascal_bornet @HeinzVHoenen @SpirosMargaris @richardturrin @Xbond49 @psb_dc @rshevlin @JimMarous @IanLJones98 @Khulood_Almani @enilev @GlenGilmore @DeepLearn007 @KamLardi @debashis_dutta @sallyeaves @EstelaMandela @NevilleGaunt @IngridVasiliu @Eli_Krumova @baski_LA

939