# Five Big Bets for AI Engineering in 2026
## Introduction: Planning the Future
Every plan is a bet on the future. Every bet carries risks and rewards. The trick isn't execution—it's choosing the right bets in the first place. Those who plan the future tend to create it.
What follows are five big bets for 2026, pointing to where agents, time, money, and compute should be directed. Some of these bets are expected; others will surprise. This framework can serve as a roadmap for any engineer navigating the rapidly evolving landscape of AI-powered development. Midway through, there's a major announcement that ties it all together—something built specifically for engineers who understand that the future of engineering is agent-first.
## Bet One: Custom Agents
The first big bet is straightforward, with clear and massively untapped ROI: custom agents. If you're working with a single agent through an agentic coding tool like Claude Code or similar CLI interfaces, you're already ahead of the majority of engineers. The 80/20 of software engineering can be summarized in one word: agents.
But if you're using custom agents in production systems, you're immediately in the top 1% of engineers. Why does this distinction matter so profoundly? Because custom agents enable scaled, domain-specific engineering work in fractions of the time and complexity it once required. When you encode solutions into custom agents, you gain repeatable problem-solving with compute.
**The Risks:** Building effective agents takes time and practice. The learning curve is steep, and over-reliance on AI will only increase. This is how technology works—it gives and it takes.
**The Rewards:** This puts you in the top 1% of engineering output. Custom agents deployed in production represent the defining competitive advantage of 2026. The journey involves building hundreds of imperfect custom agents before arriving at truly great ones.
## Bet Two: Deprecating Old Engineering Skills
Sometimes moving forward means leaving things behind. In 2026, we must deprecate our old engineering skills to make space for new ones.
Think of a pencil as a symbolic visual—the perfectly balanced tool for builders. Creation exists at one end, deletion at the other. You're constantly drafting new skills and erasing what no longer serves you. Over time, both ends wear out, and you need to pick up a new tool. The pencil reminds us that growth isn't just about writing the next line, but knowing when to erase and start again.
There are things you're doing right now that your agent could absolutely do for you. The critical shift in the age of agents is understanding it's no longer about what you can do—it's about what you can teach your agents to do. This represents a fundamental differentiating factor in how we approach work.
**The Risks:** Breaking multi-year, multi-decade habits will be uncomfortable. Skill atrophy isn't potential—it's guaranteed. Think about calculators, GPS, Google Maps. These technologies deprecated our abilities to do what they now handle automatically.
**The Rewards:** The productivity gains and future-proofing of your engineering career will be massive when you let go and stop doing things the old way. Many engineers know that agents could handle their work, yet they still type line by line, still copy and paste, still write prompts they know could be automated into reusable agentic workflows. They continue doing things the slow way because it's how they've always done them. We must deprecate old skills to make space and time for new ones.
## Bet Three: Limit Breaking the Agentic Coding Constraints
If you're building valuable work with agents, you've likely encountered the two constraints of agentic coding: planning and reviewing.
We can now build new plans and hand them off to agents at higher rates than ever. But our involvement in the planning process takes time. At the end of the loop, when agents finish their work, reviewing takes the most time. These are the constraints of agentic coding, and in 2026, we're going to blast through them—limit breaking these constraints.
The answer lies in custom agents, but not just standalone agents. The most powerful abstraction for engineers right now is custom agents inside ADWs (Agent Development Workflows)—a pipeline of agents that combines the old world of engineering with the new world of agents. The goal is breaking through these constraints to unlock insane scale.
**The Risks:** Implementation here is complex. The upfront investment will cost. This is where casual coders will fall away. You must have deep understanding of your codebase to increase your planning and reviewing velocity. You cannot teach something to do what you yourself don't know how to do.
**The Rewards:** There's an estimated one to two orders of magnitude improvement lying at the end of this journey if we can successfully limit break the planning and reviewing constraints of agentic coding.
## Bet Four: Multi-Agent UIs and Voice Interfaces
The chat interface is the laziest and most overused interface for agents. In 2026, the focus shifts toward multi-agent UIs and interaction interfaces—specifically voice.
The whole point is increasing the information rate between you and your agents. You want more work done faster, commanding more compute with multi-agent UIs and faster interaction interfaces. This lets us wield more compute faster than ever. If you could communicate everything in your head right now to a multi-agent system that can do the work for you, wouldn't that fundamentally change your future? These systems are right around the corner, and we can build them now.
**The Risks:** User confusion represents a major concern. Putting multiple agents into a single UI and orchestrating all that feedback, logging, and critical information flow is challenging. This presents a complex UI/UX design challenge that requires both technical skill and thoughtful design.
**The Rewards:** The right multi-agent system could yield another order of magnitude increase in engineering output. We're talking about use-case-specific multi-agent systems and UIs—not just another wrapper or background task manager. This is about getting products to have many agents operating on them, accomplishing more work faster than ever. Everyone's going too wide. It's time to focus, go deep, and solve real engineering problems with this incredible technology.
## Bet Five: Compute Maxing
Here's a secret: however much compute you think you're using, however skilled you believe you are at agentic coding—you're not using that much compute, and you're not that great yet.
Unless you have an always-on, 24/7 agent solving problems on your behalf all the time, you can and should be using more compute. Compute maxing will be a major focus for 2026, and it's highly recommended for every engineer.
The answer isn't jumping immediately from prompting back and forth with coding tools to suddenly building a 24/7 agent. We're engineers—we operate in the gray, the messy, the progressive nature of it all. We should always be thinking: How can I use more compute today?
Consider all the developer workflows, products, tools, and applications that remain untouched by valuable language models, prompt chains, and agents. There's tremendous potential waiting to be unlocked.
**The Risks:** Operating 24/7 and doing real engineering work will cost money. There's a side bet that compute costs will continue declining, and we'll continue getting great open-source, low-cost, high-intelligence models. But still, there will be architecture and infrastructure complexity, and we'll probably waste significant compute getting there. You need to put in the repetitions to even know what's possible.
**The Rewards:** If you're compute maxing, you have living software that works for you while you sleep. This has been the mission all along: build living software that works for us while we sleep. This represents true agentic automation and exponential leverage. Imagine these systems always on. Recent releases like OpenAI's ChatGPT Pulse—with the tagline "Now ChatGPT can start the conversation"—point toward this future. There's an earlier prediction worth remembering: the "two-way prompt," where these systems become so powerful they start prompting us. At some point, the conversation will flip, and our agents will prompt us more than we prompt them.
## The Unified Vision: Tactical Agentic Coding
These five big bets—custom agents, deprecating old skills, limit breaking agentic coding constraints, multi-agent user interfaces with voice, and compute maxing—all lead to something transformative. To scale your impact, you scale your compute. This is the equation for winning engineering in 2025 and beyond.
The agent architecture has changed everything, creating a massive opportunity: **Tactical Agentic Coding**.
This represents phase two of agentic coding education, built for engineers in the AI age. It's designed to help you master the tactics of agentic coding to scale far beyond AI coding and casual experimentation, achieving advanced engineering so powerful that your codebase begins to run itself.
### What Is Tactical Agentic Coding?
This is for engineers who ship. We know that AI coding alone isn't enough. Several big ideas that engineers are missing right now represent opportunities to get ahead:
- AI coding was just the beginning
- Casual coding represents the lowest-hanging fruit
- 95-99% of all codebases are now outdated and inefficient
- You, the engineer, are the bottleneck—not the models, tools, or agents
- Debates about the best agentic coding tool, model, or pricing miss the big picture
- It's a skill issue now
- A new engineering role has emerged: the agentic engineer
This isn't about models, prompts, or individual agents. It's about composing the old world of software engineering with the new world of agents to achieve exponential scales of engineering.
The course asks one guiding question: **What if your codebase could ship itself?**
### Who This Is For
This is not for beginners. If you're not shipping to production, not a mid-to-senior-level engineer, or not working with agent coding tools, this isn't the right fit. The mission is to transform engineers into professionals they can't replace—accomplished through two concrete actions:
1. Let go of the old ways of engineering
2. Master the new best tool for the job
By internalizing the eight tactics of agentic coding, you become an engineer who builds the system that builds the system. It's no longer about what we can do—it's about what we can teach our agents to do. We're building the system that builds the system.
There will be an engineer on your team—hopefully you—orchestrating fleets of agents inside agentic systems that handle all the application work you're currently spending time on. Agents represent how you win the game of engineering in 2025 and beyond.
### Course Structure
The course contains eight lessons with eight tactics, progressing from beginner to intermediate to advanced levels:
**Beginner Lessons:** Foundation-setting, though lesson one may feel slow to experienced engineers—but this groundwork is critical for everything that follows.
**Intermediate Lessons:** Turn your prompts into plans, and your plans into next-generation agentic prompting techniques. AFK agents close the loops. Scale your compute to gain confidence. Build self-correcting systems that destroy bugs and hallucinations.
**Advanced Lessons:** Let your agent focus. Implement agentic review and documentation. A big secret reveal in lesson seven. And the one big idea to rule them all: the agentic layer.
### Agentic Horizon
Extended lessons go beyond the core curriculum for those wanting to push further. These include three current lessons that have been teased over previous weeks, plus three future lessons that students get to vote on—a new voting mechanism that lets you decide exactly what comes next.
### Investment and Access
Early bird specials reward engineers who move fast and invest in themselves. Members of previous courses receive significant discounts. Team purchases are available with scaling discounts for multiple seats.
The offering is completely risk-free with a no-questions-asked 30-day refund policy before starting lesson four. If the style, ideas, or vision don't resonate by lesson three, you get everything back. The goal is having the right engineers in the course.
## Conclusion: Your Plan for 2026
No matter what path you choose, focus on your plan for the future. The five bets outlined here represent one framework:
1. **Custom Agents:** The foundation of competitive advantage
2. **Letting Go of Old Skills:** Making space for new capabilities
3. **Limit Breaking Review and Planning Constraints:** Unlocking order-of-magnitude improvements
4. **Multi-Agent UIs:** Using more compute than ever through better interfaces
5. **Compute Maxing:** Always asking how to use more compute
Whatever you're doing right now, however much compute you're using—you're not using enough. There's a prediction for 2026: we'll see our first compute crunch, a limited supply of compute. We're already seeing signs. As engineers on the bleeding edge scale up compute usage, this will become a larger problem. The more compute you can use, the more you realize you can do, which leads to using even more compute.
This is the plan for 2026. The opportunity is massive and available now. No matter what you decide, stay focused and keep building.
