OpenAI's 2028 AI Research Goal & Cheap Chinese Models: What Builders Need to Know

OpenAI sets a 2028 target for automated AI research. We break down what this means for founders, product leaders, and engineers. Plus, hear why cheap Chinese models like DeepSeek and 01.ai are changing AI economics, Apple's privacy-first Siri, and the talent wars heating up.

Transcript

SPEAKER_00 0:00

OpenAI just put a date on automated AI research. We're talking March 2028. And Chinese models are now dramatically cheaper, shifting the entire economics of building with AI. Also, Apple just raised the bar on context-aware assistance directly impacting how you'll design your next product. 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 that hit this week. First up, OpenAI unveiled its phase 3 roadmap, targeting an automated AI researcher by March 2028 and laying out a vision for a personal AGI assistant for every human. This isn't just a vague aspiration anymore. OpenAI published this plan right alongside its confidential IPO filing, giving it some serious weight. Essentially, they're talking about systems that can develop, test, and even deploy new models autonomously, though they emphasized human oversight is still crucial. For builders, this is a massive signal. It means we need to start thinking about AGI not as some far-off sci-fi concept, but as a consumer product with a tangible timeline. Are you ready for agentic systems that can build other AIs? And how will a deeply personal, privacy-centric AGI assistant shift everything from UI design to data integration in your products? It's a complete rethink of the stack. Next, we saw a significant talent shift. John Jumper, the lead for Alpha Fold and a Nobel laureate, has exited DeepMind to join Anthropic. This is a big deal, signaling a shifting center of gravity in the core AI research space. Jumper is a heavyweight and his move to Anthropic, a key competitor in the frontier model race, is a notable gain for them and arguably a loss for Google DeepMind, from which this is reportedly the third senior exit in three months. What's more, DeepMind's demis Hassabis recently suggested that AGI could arrive around 2030, plus or minus a year. For us builders, these aren't just academic curiosities. These timelines and the movement of top talent directly inform our multi-year product roadmaps, hiring strategies, and where we place our bets on future platform shifts. Which labs are gaining momentum, which timelines feel more credible for your long-term planning. Also, making waves is a drastic change in model economics, especially from China. Deep Seek reportedly trained its V3 model for just $5.58 million and R1 for an astonishing $294,000. This isn't just a small discount. Their V4 Pro pricing was slashed by 75% in May, and Zero Bond AI is now advertising inference at about 0.14 per million tokens. That's an order of magnitude shift. The drivers behind this are fascinating too. They're citing advancements like sparse mixture of experts, MOE architectures, and FPA training, which are making these models incredibly efficient. What does this mean for you? Well, these cost shifts enable entirely new applications and business models, especially for markets where cost sensitivity is king. Can you now afford to embed sophisticated AI into every corner of your product? It's time to evaluate these Chinese models and consider their trade-offs against Western alternatives because the price point could unlock entirely new product lines. Moving on, Apple has shipped its Siri AI developer beta, showcasing a deeply context-aware on-device assistant powered by private cloud compute. This new Siri, leveraging Apple Intelligence and Gemini, can read and understand context from your messages, photos, and emails, and then integrate that understanding deeply with iOS apps. Imagine Siri actually scheduling an event based on a conversation in your messages or summarizing emails for you all while knowing what's on your calendar. It's a massive leap in capability, but crucially it's designed with privacy at its core, using private cloud compute for processing without persistent data storage. For builders, this raises the bar significantly for all context-aware assistants. You'll need to plan app integrations and data handling patterns that assume local context and minimal data egress. The user expectation for intelligent private assistance just skyrocketed. And finally, we're seeing the emergence of powerful open source alternatives. Magnitude, a new open source coding agent, was released running on models like DeepSeqv4Flash and GLM 5.2. This agent is pitched as a cost-effective alternative to proprietary tools like Copilot, and critically, it supports local privacy-preserving workflows. This isn't just another GitHub project, it's a demonstration that viable, high-quality coding agents can be built and run on open models without relying on expensive locked-in APIs. For product leaders and engineers, this signals an acceleration in the open source agent ecosystem. Expect more tools like this, offering greater flexibility, reduced vendor lock-in, and potentially massive cost savings on developer tooling. It's an exciting time to be building with open source AI. Now, if I had to pick the single most important story from this batch, it's got to be OpenAI's phase three roadmap and its implications for builders racing toward AGI. What happened here? OpenAI, traditionally a bit cagey about hard timelines for AGI, just went public with a very specific dated plan, an automated AI researcher by March 2028. This isn't AGI itself, but it's a critical stepping stone. An AI that can do AI research and development. This roadmap, which also includes accelerating global economic growth through productivity and a personal AGI assistant for every human, came out right as they're reportedly filing for a confidential IPO. It's a statement. Why does this matter right now? Because it transforms AGI from a nebulous academic ambition into a concrete strategic target with a deadline. This isn't just about research, it's about shifting market dynamics. Investors, talent, and other big players will be aligning their bets based on these timelines. If an AI researcher can automate parts of model development by 2028, that puts immense pressure on anyone competing solely on base model capability. The value will rapidly shift upstream to application integration and user experience. So who should really care about this? IT founders. If you're building a startup in the AI space, you need to ask how your defensibility holds up when model development accelerates and becomes potentially commoditized. Your advantage will be in unique data, vertical applications, and distribution, not necessarily just having a slightly better model. While the core model training might become automated, the demand for robust, scalable infrastructure for deployment, fine-tuning, data management, and privacy preserving compute will explode. Value shifts from building the core model to enabling its safe, efficient, and ethical deployment at scale. Ah, they asked indie hackers. This presents a fascinating opportunity. If the base tools for AI research become more accessible and automated, could you leverage these to build highly specialized agents for niche problems faster and cheaper than ever before? How I'd think about it as a builder, I'd view this as a clear signal that the AI development cycle is about to accelerate exponentially. Imagine a factory that not only produces cars but also designs and builds better car factories. That's the implication of an automated AI researcher. The strategic play isn't to compete with that factory, but to figure out what new kinds of transportation experiences you can create once car production becomes super efficient and personalized. It means focusing on the human AI interface, the ethical guardrails, and the deep vertical problems that even advanced general AI will need help solving. My nobiest take on this is that it's a powerful signal, but we need to read the fine print. While automated AI researcher by 2028 is a bold target, it's not synonymous with AGI itself, and personal AGI still lacks product specifics. The IPO context also means there's a strong incentive to paint an optimistic picture. DeepMinds Demish Hasabi's ADA 2030 timeline is a good counterpoint. It's a credible aspiration that sets a demanding pace for the industry, but it's a goal, not a guarantee. Plan for it, but don't bet your entire company on its exact arrival date. If you want one practical takeaway from today's episode, here it is. Experiment. Evaluate Chinese models for cost-sensitive paths. Here's how to try it in under 60 minutes. 1. Pick one non-critical workflow in your team or even a side project. Think about internal processes like summarizing long meeting notes, generating first drafts for internal communications, creating simple code snippets, or automating basic customer support responses, where accuracy can be slightly more forgiving than a core product feature. 2. Run DeepSeek V4 Flash or 01.ai in parallel with your current model. Most of these models offer straightforward API access. Set up a quick script to send the same prompt to your existing model, EGCAR, GPT-4, Claude, and one of these Chinese alternatives. You're not replacing anything yet, just observing. 3. Compare latency, accuracy, and critically the cost. Log the time taken for responses, a simple qualitative assessment of output quality for your specific task, and then tally up the token usage and the resulting cost. Pay close attention to where you see those reported 10x or even 100x cost savings. Why is this specific experiment worth your time right now? Because the economics of AI are fundamentally shifting. These models might not be the absolute best for every task, but if they can deliver good enough results at a fraction of the cost, we're talking orders of magnitude cheaper, they could unlock entirely new product features, expand your addressable market by allowing lower price points or dramatically improve your margins. It's about finding the sweet spot where you can leverage these massive cost efficiencies without compromising your core value proposition. Don't assume your current model stack is the only viable one. 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.