No‑BS AI Briefing
No‑BS AI Briefing is for builders who don’t have time for hype. Each episode focuses on a handful of high‑signal stories in AI and AGI, unpacked in simple language with a builder’s perspective. You’ll hear what changed, why it matters, and how you can experiment with the tools, ideas, or strategies yourself—whether you’re leading a team, shipping a startup, or exploring AI side projects.
No‑BS AI Briefing
OpenAI GPT-5.6 Sol Tiers, EU AI Act, & Agentic Workflows for Builders
Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.
OpenAI just dropped GPT 5.6 solved with new agentic reasoning modes and a tiered pricing structure. Meanwhile, the EU has locked in its AI Act Enforcement deadline for August 2nd, and we're seeing persistent AI agents arriving directly in tools like Slack. No BS AI briefing brought to you by ProActive AI. Welcome back. I'm your host, Vikash Sharma, and this is where builders get straightforward AI news without the fluff. Alright, let's dive into some high signal items this week that I think really matter for anyone building with AI. First up, we've got big news from OpenAI. They've launched GPT 5.6 Sol, and this isn't just a bump in version numbers, it's a significant architectural shift. In plain English, OpenAI unveiled GPT 5.6 Sol with a couple of variants, Terra and Luna. The big news is that Sol itself adds a new max reasoning mode and even more interestingly, an Ultra mode that uses subagents. So instead of one monolithic model trying to do everything, it can now break down tasks and coordinate smaller AI subagents to tackle complex problems. This model also sets a new state of the art on Terminal Bench 2.1, which focuses on command line workflows and shows improvements in specific domains like biology and cybersecurity tasks. For builders, this is huge. That agentic architecture using subagents means you can now envision and build far more complex product automations without needing to orchestrate every step yourself. Think about multi-step data processing or intricate customer support flows. Plus, the tiered pricing both Seoul at $5 or $30 per million dollar tokens, Terra at $250 or $15, and Luna at $1.6 really lets you match the cost and performance to the specific task. Don't need the ultra mode for a simple summarization. Use Luna and save a ton. And if that wasn't enough, OpenAI also announced a Cerebras hardware launch coming in July, targeting an impressive 750 tokens per second throughput. This signals that they're getting serious about infrastructure readiness for high volume real-world deployments. So from complex agentic workflows to cost-optimized tiers and serious throughput, there's a lot to unpack here. Next, GitHub is also making big moves, releasing MI Code 1Flash, and a new copilot SDK specifically for enterprise agents. What happened here is that MI Code 1Flash is now generally available for copilot business and enterprise users. This model is engineered for speed and enterprise grade reliability, meaning your developers can expect faster, more consistent code suggestions and completions, especially within iterative agent workflows. But perhaps even more impactful for builders is the integration into GitHub Desktop 3.6, which now incorporates Copilot for tasks like crafting commit messages or resolving those annoying merge conflicts. This comes via a new SDK that offers crucial capabilities like model selection and critically bring your own key or BYOK. This is a game changer for enterprises that need specific controls over their data and model usage. The new enterprise settings also include strict known marketplaces, which lets organizations restrict the plugins developers can use, further enhancing security and compliance. So, what does this mean for you, the builder? You're getting an enterprise optimized coding model that's built for speed in your agent-driven workflows. That unified SDK simplifies the process of embedding multi-model agents directly into your existing development tools and flows, making them more seamless. And for engineering leaders, those new governance controls are invaluable for improving your production security posture and managing model access, which we all know is becoming increasingly vital. Also, Anthropic is shaking things up in the collaboration space by launching Claude Tag, their new Slack native persistent AI agent. This is pretty cool. The beta of Claude Tag introduces an AI teammate that basically lives in your shared Slack channels. Unlike a chatbot that you invoke for a single interaction, Claude Tag maintains context across conversations, can run multi-step tasks using various tools, and then posts its results directly into relevant threads. It even has an optional ambient mode, so it's always listening, always ready to assist, without you having to explicitly ping it every time. It's available for Cloud Enterprise and Team customers. For builders, this is a significant step towards truly persistent context-aware automation right inside your team's communication hub. Imagine an AI that tracks project progress, summarizes long discussions, or proactively pulls up relevant documentation based on ongoing conversations. It's a turnkey deployment too, which means you don't need to build custom integrations to get this AI assistant working within your Slack ecosystem. Ultimately, this type of always-on agent can drastically reduce context switching for your team and significantly boost overall efficiency, freeing up human team members for higher value work. Now, shifting gears to policy, the EU AI Act Enforcement Deadline has been confirmed for August 2, 2026. Law 360 reports that EU institutions have reached a provisional omnibus agreement that addresses many of the implementation challenges previously discussed. The full obligations for high-risk AI systems will apply from that date. Why does this matter for builders? Well, it sets a fixed concrete compliance deadline, which means you can no longer kick the can down the road. If you're building high-risk AI systems that will operate in the EU market, you now have a clear timeline to get your house in order. This enables concrete planning for everything from risk assessments to detailed documentation, transparency requirements, and establishing robust human oversight mechanisms. For any startup or enterprise targeting the EU market, compliance isn't optional, it's a hard requirement. Non-compliance could risk substantial penalties, so getting this right is crucial for market access and protecting your business. And finally, something truly inspiring and a fantastic use case of AI. AI has unlocked the first entire Herculaneum scroll. This comes from the Vesuvius Challenge. A team successfully raid an entire scroll Perk 1667 scroll 4 without ever having to unroll it. Imagine that. A scroll carbonized over 2000 years ago by the eruption of Vesuvius. They achieved this by using ESRF X-ray scans combined with machine learning techniques to detect and extract the incredibly faint ink from the carbonized papyrus. It's essentially like reading a book that's been burnt to a crisp from the inside out. What's even better for builders is that all the data and code from this project have been released under open licenses. Why does this matter for you? It's a powerful demonstration of how AI can solve incredibly complex multi-step problems where the signal is extremely low and obscured by noise. This isn't just about ancient scrolls, it's a blueprint for similar applications in fields like advanced medical imaging, detecting tiny anomalies in scans, or in material science, understanding internal structures without destruction. And because the assets are open, other builders can replicate the process, extend it, and adapt it to new challenges. It's real-world impact that's tangible and it goes far beyond just generating text or images. Pretty incredible, isn't it? Alright. For our deep dive today, I want to zero in on that OpenAI GPT 5.6 Sol launch, specifically focusing on this new agentic model tier system and its broader product impact. This, to me, is the most important story in the batch because it really signals a profound shift in how we'll be building with AI. We're moving towards sophisticated multi-model architectures where builders have to consciously match the capability and cost of the AI to the specific task at hand. What happened is that OpenAI didn't just release a faster, smarter model. They essentially rolled out an entire new product line within GPT 5.6 Sol. You've got Sol itself with its advanced max reasoning mode, but then the truly groundbreaking part is the Ultra mode, which internally uses subagents to break down and solve problems. Alongside this, they've introduced Terra and Luna as progressively lighter, more cost-effective options. This tiered system, paired with the promise of 750 tokens per second throughput from Cerebrus hardware and a clear pricing structure, tells us they're thinking about the full deployment lifecycle for complex AI applications. Why does this matter right now? It's because the market for AI is maturing beyond just simple API calls to a single monolithic model. As builders, we've often had to choose between a powerful, expensive model or a cheaper, less capable one. Now, OpenAI is giving us granular control. This tiered pricing allows for incredible cost optimization. Why pay for a Ferrari when a scooter will do for a quick trip? But more importantly, the Ultra Mode with subagents is a clear nod to the agentic future. It's saying don't just prompt orchestrate. This is about building products that can autonomously handle complex, multi-step tasks, reducing the manual overhead in your applications and potentially opening up entirely new use cases. The Cerebras integration also matters for scale. It means you can actually deploy these complex, multi-tiered agentic solutions at a significant volume without worrying about hitting performance bottlenecks. So who should really care about this? Definitely Vertil Founders. You need to be thinking about how these cost tiers affect your unit economics and how the agentic capabilities can power entirely new product features or even new business models for TeddyG product managers. This means thinking beyond simple chat interfaces and envisioning products where AI takes on a more proactive, persistent role managing workflows and delivering more complete outcomes. Infrastructure engineers. Need to pay close attention to the Cerebrus announcement. Planning for throughput and efficient resource allocation across these different model tiers will be crucial for managing costs and ensuring reliability. And frankly, our indie hackers should be thrilled because these lower cost Luna and Terra models open up opportunities for building and experimenting with AI at a much more accessible price point, even for those complex ideas that might leverage Soul's Ultra Mode when needed. How I think about it as a builder is to start treating model selection like a core product decision. It's no longer just about picking GPT-4 or Claude, it's about creating a routing layer in your application. Think of it like a smart traffic controller. For routine summarization or classification, you route to Luna low cost, good enough. For complex code generation or detailed research, you send it to SOL's max mode. And for truly multi-step autonomous tasks like filling out forms or executing a sequence of API calls, you leverage Soul's ultra subagent mode. This isn't just about saving money, it's about building more robust, intelligent, and adaptable systems that can automatically scale their intelligence based on the complexity of the task. It's like having a team of specialized AI workers where you assign the right person to the right job instantly. This level of control and capability wasn't really practical at this scale just a year ago. My no BS take on this is simple. This isn't just marketing hype. OpenAI is clearly pushing the boundary on agentic capabilities while simultaneously addressing the very real builder need for cost control and deployment flexibility. While benchmark generalization always needs to be validated against your real workloads, the intent here is clear. They want to enable more complex autonomous AI products. Don't get lost in the raw numbers, focus on the architectural shift and the implications for how you design your next AI-powered feature. If you want one practical takeaway from today's episode that you can act on this week, here it is. Plan a multi-model strategy for your product or internal workflows. This is more than just a thought exercise. It's about making concrete decisions about how you leverage these new capabilities and pricing tiers. Here's how to try it in under 60 minutes. First, yeah. Map out two to three existing product workflows or internal tasks that currently use an LLM. Think about things like content generation, data extraction, customer support routing, or internal knowledge base queries. Break each of these down into their constituent subtasks if they're complex. Second, for each subtask, assign the minimum viable model tier, e.g. D, Luna, Terra, or Sol that could reliably complete it. Be honest here. Does a simple sentiment analysis really need Sol's ultra mode? Probably not. Can a basic summarization be handled by Luna? Likely. Identify where Sol's max or ultra mode is genuinely necessary for its advanced reasoning or agentic capabilities. Third, estimate the cost deltas for routing each subtask to its assigned tier versus using a single higher tier model for everything. You don't need exact figures, rough percentages are fine. This quick cost benefit analysis will highlight potential savings. And finally, prototype one of these routed workflows. Pick a simple one where you identified a clear tiering opportunity, even if it's just a few lines of code to call a different API endpoint based on task complexity, build it out, measure the latency, and observe the output quality. Why is this specific experiment worth your time right now? Because the era of one model fits all is rapidly ending. This move by OpenAI and likely others to follow means that optimizing your model usage isn't just about efficiency, it's a strategic decision that impacts your product's performance, cost effectiveness, and your ability to build truly intelligent agentic features. By proactively planning a multi-model strategy, you're not just saving money, you're building future-proof, adaptable AI applications that are ready for the next wave of innovation. Seriously try this. That's it for today's NoBS AI briefing. If this helped, follow the show in your podcast app and share it with one builder you know. And if you've got questions or topics you want covered, connect with me on LinkedIn and send them over. See you in the next briefing.
