🏗️ 🚢 🚀 Tinker, Agent Memory, and Prompt Optimization; Thoughts on AI Engineering


Hey Practitioners,

I hope that your new year is off to a great start, and that you took some time over the break and into 2026 to outline your specific AI Engineering career goals.

It's getting wild out there for everybody, and you know what they say about not having a plan 💀.

📈 By all accounts, it's going to be a great year for us AI Engineers and AI Engineering Leaders. According to LinkedIn, the AIE is the fastest-growing job out there. Interestingly, the number 2 fastest-growing job is "AI consultants and strategists."

TL;DR: The most important (and valuable) set of skills out there is to know how to build agentic AI systems, but the second most important skill is to understand the business problems they should be solving, the why behind building the agents.

It's safe to say that if you can do both, you'll be unstoppable 🦾.

I believe this is true for individuals and companies. The ones who will win in the marketplace in 2026 can answer two important questions:

  • "What agents should we be building and why?"
  • "Given that, how should we prototype our agents, get it into the hands of our users, and ship it to production with code that will scale with usage?"

It turns out that the best AI Engineers - the unicorns 🦄, I suppose you might say - can, of course, do both. They can span the spectrum between Product, Engineering, and Data Science, as Agent Engineering requires. Said differently, they can play the role of AI Engineer or of "AI Scientist," a term coined (most recently, at least) by Certified AI Engineer Jared Rand this week.

The AI Engineering space is challenging to navigate. The consultants, scientists, researchers, and product managers are filling engineering skills gaps, but at the same time, software engineers are learning data science and product management!

AI enables us to do more. So, we'll have to do more.

But what should we do?

Here's how I like to think about it, and how I've been recommending to people and companies that they might think about: don't ask what to do. First, ask what not to do.

Are you an individual considering your career? Ask "what do I definitely not want to do most of the time?" If you don't love coding, then learn enough to vibe code an agent with something like LangSmith's Agent Builder. If you don't love product management, then at least be able to brainstorm a use case using something like ChatGPT Use Cases for Work, and then use a reasoning model to help you write a plan for it. If you're not obsessed with defining the next evolution of the data scientist like Jared, at least be able to tinker with some automated fine-tuning. If you're personally struggling with your journey, feel free to reach out to Coach Mark, our new Student Success Manager, or to me directly, and we'll give you our best advice for your situation.

Are you an enterprise considering your 2026 strategy? Ask "of the many ideas we've come up with so far, which ones am I not sure will work?" Nix those ideas for now to remove unnecessary "nondeterministic" technical risk and go for low-hanging fruit. Focus on quick wins to build momentum by solving problems that people already solve well. Ask "of our current projects, which ones will definitely not require agents?" Then focus on triaging the other ones! If you're looking for more customized support for your enterprise, reach out to our consulting team about custom solutions.

In summary, there's never been a better time to jump into a new field and get your hands dirty by building 🏗️, shipping 🚢, and sharing 🚀 in an AI-enabled way. If you do this, you can stay on offense while focusing on all of the opportunities - the new jobs and many problems to solve - rather than worrying about all of the potential downside risks.

Until next month!

Cheers,

Dr. Greg

LIVE TOMORROW!

Recursive Language Models

Join us live tomorrow at 1:00 PM ET for RLMs! Are they truly THE paradigm of 2026? Do we really need them for context engineering in 2026? How do they help solve problems like Context Rot, and why are they better than other Context Folding approaches? Join us to learn everything you need to know about RLMs as an AI Engineer today!

NEXT WEEK!

Ralph Wiggum: Continuous Agent Loops in Claude Code

What the heck is the deal with Ralph, the Ralph Wiggum phenomenon, and Claude Code? In short, it's all about agent loops and iterative improvement. Join us live for the full first-principles breakdown of what's going on, what you need to know about the Ralph development methodology, and where it should fit into your AI Engineering toolkit.

IN CASE YOU MISSED IT!

Agent Memory

Remembering stuff is a simple idea, but how do we implement memory in our AI applications, exactly? We learned how to build short and long-term memory, how memory intersects context engineering, and when we should leverage memory (spoiler: all the time).

Tinker: a Fine-Tuning API

In this one, we dove into the Tinker API, from Thinking Machines, right before they hit the news cycle! The API is dope though, it (classically!) uses LoRA, which we dove into, and it also does RL quite nicely, as we demo'd. It's a great tool for beginners to pick up, as mentioned up top in today's newsletter!

Prompt Optimization

We dove into the latest in prompt optimization, which of course meant DSPy, again! From meta prompting to GEPA (Genetic-Pareto), we dug into how we can get the most out of our prompts today by "optimizing across the agent harness" and getting directed feedback beyond simple metrics.

The AI Engineering Bootcamp v1.0

The AI Engineering Bootcamp, Cohort 10, kicks off on April 7th. We also just rolled out a new discount policy for community members! Early bird pricing has begun - get up to 40% off with a Certified Referral!

To keep learning, check out the Awesome AIM Index, YouTube, or our open-source courses, including LLM Foundations, LLM Engineering, and LLM Ops.

Keep building 🏗️ shipping 🚢 and sharing 🚀, and we'll do the same!

Cheers,

Dr. Greg, The Wiz, Jacops, Laura The Legend, and all of the AI Makerspace staff, ambassadors, and volunteers we couldn't run our courses and community without!