The AI Daily Brief: Artificial Intelligence News and Analysis

Episode: How AI Starts Doing the Work in 2026 With Anthropic CPO Mike Krieger
Host: Nathaniel Whittemore
Guest: Mike Krieger, Chief Product Officer of Anthropic
Date: December 24, 2025

Episode Overview

In this end-of-year episode, Nathaniel Whittemore (“NLW”) sits down with Mike Krieger, co-founder of Instagram and Artifact, and current CPO at Anthropic, to explore how AI’s role in coding, work, and enterprise is rapidly evolving. The conversation zeroes in on the emergence of agentic and “vibe coding,” the changing landscape for both enterprises and individual users, new interaction patterns with AI agents, and predictions for how 2026 could mark a tipping point where AI begins to reliably “do the work.”

Key Discussion Points & Insights

1. Anthropic’s Early Focus on Coding & Agentic AI

• Clarifying Early Intentions

  • Krieger emphasizes that Anthropic always believed coding would be central to AI’s utility, not just for software creation but as a problem-solving tool ([02:05]).
  • “...For very powerful AI, you need the ability of the model to reason about things, to plan genetically and work for a long time horizon, but then also to be able to write and run code…” — Mike Krieger ([02:05])

• The Claude Model’s Milestones

  • The release of Claude 3 marked a turning point when the outside world realized these models could generate substantive code, beyond basic functions ([02:48]).
  • Anthropic’s artifact product was among the first to enable interactive, AI-driven coding experiences outside a formal IDE ([03:01]).

2. The 2025 ‘Year of Agents’—Especially Coding Agents

• Evolving Use Cases and Community Adoption

  • 2025 saw “coding agents” quickly become the most exciting use case, confirming internal bets at Anthropic ([04:39]).
  • Internal hackathons revealed a breadth of applications – from bioinformatics to “SRE-in-a-box” – demonstrating agent tech’s applicability far beyond just engineering teams ([08:01]).

• Designing for Non-Existent Capabilities

  • Krieger discusses Anthropic’s product principle “ride the exponential.”
    • The team builds products for today’s needs but ensures they will naturally improve as models do ([06:42]).
    • Sometimes improving the product meant removing scaffolding as models became more capable ([06:42]).

3. Expanding Accessibility: From Engineers to Tinkerers & Non-Technical Users

• Non-traditional Users Adopting AI Tools

  • Internal and external hackathons showed non-engineers quickly using these tools in creative ways ([08:01]).
  • Anthropic rebranded “Claude Code” to the “Claude Agent SDK” to better reflect the diverse array of tasks agents tackled ([09:20]).

• The Emerging Gap: Usability and Trust

  • Current tools still most benefit “tinkerers” who bridge technical and non-technical roles ([10:04]).
  • The challenge now is making these systems both more understandable and reliable for true non-developers.
    • “If you had a human coworker…sometimes it would sort of make a mistake you would never have expected…it'd be a pretty complicated relationship with that coworker.” — Mike Krieger ([10:04])

• Changing Habits and Patterns of Work

  • Even power users are only just beginning to truly rewire their workflows for “vibe coding” ([11:26]).
  • Krieger shares a personal story: Once he’d built one tool in Replit using Opus, it became second nature to build more (even automating a Secret Santa for his family over breakfast) ([12:17]).
  • “I do think…there's this sort of still like habit creation and adaptation of even knowing you can do that that we still need to close.” — Mike Krieger ([12:17])

4. The “Three Buckets” of AI Coding Adoption

• Software Engineers:

  • Fast adopters, iterative feedback cycles, push the limits of agent capabilities ([18:34]).

• Vibe Coding / Builders / Product Managers:

  • Need help climbing the complexity curve; often stymied by “magic words” or knowledge gaps ([18:34]).

• Enterprise / Business Users:

  • Focused on productivity, want high-quality outputs—not just new interfaces or chatbots ([21:25]).

• Bridging the Gaps

  • AI tools must help non-engineers “move up that complexity ladder in a structured, thoughtful way” ([18:34]).
  • Krieger draws an analogy to Instagram’s own early growth pains—a product may initially be simple, but true value comes as it scales and infrastructure matures ([18:34]).

5. Enterprise AI in 2026: Infrastructure, Reliability, Integration

• Big Trends for Enterprises

  • Growing focus on deploying “horizontal agents”—copilot-style AIs that can handle both creative and repetitive back-office tasks ([22:16]).
  • Enterprises move from “AI sprinkles” (chatbots, isolated features) to agent-native workflows that rethink core product or service design ([22:16], [23:55]).

• Infrastructure, Retraining, & Data Readiness

  • 2026 may become an “infrastructure year” for enterprises, focusing on making systems “AI-friendly” through better data annotation, process mapping, and robust connectors ([25:28]).
  • “Figuring out what are the missing connector bits is going to be, I think, a lot of 2026.” — Mike Krieger ([25:28])
  • Moving from data retrieval to AI-driven action and decision-making is the new frontier ([25:28]).

6. Roadblocks & Predictions for 2026

• Blockers:
  • Legacy systems, regulatory constraints, complex permission structures, and integrating with existing infrastructure ([27:00]).
  • Anthropic is prioritizing the “distributability” of their agent tools, making them deployable in diverse environments ([27:00]).
• From Tool to Colleague
  • 2026 may be the year when AI transitions from just a tool to something that can “reliably take work off your plate” ([30:01]).
  • Early evidence: GitHub partnership with Claude, where the agent autonomously handles pull requests ([28:42]).
  • Krieger is optimistic but realistic—total autonomy remains over the horizon but is beginning to show in specialized workflows ([28:42], [29:50]).
    • “...you might delegate to somebody else—that’s very much around the corner.” — Mike Krieger ([28:42])

Notable Quotes & Memorable Moments

• On Anthropic’s Product Philosophy:

  • “One of [our principles] is ride the exponential, which is like we're trying to build products that both, you know, meet the moment…Maybe the ones that are a little early, we won't release yet, but that they can naturally improve.” — Mike Krieger ([06:42])

• On Broadening AI Coding Beyond Engineers:

  • “When we launched it we started seeing things externally too, like people using Claude code as their project manager...as a data scientist...So we started seeing this much more, it’s why we eventually renamed the underlying SDK…because we realized calling it code was doing it a disservice…” — Mike Krieger ([09:20])

• On Changing Work Habits:

  • “It's easier to build the incremental n+1, but…that first one…requires that sort of uplift if you're not in the habit.” — Mike Krieger ([12:17])

• On the Reliability Challenge with AI Coworkers:

  • “If you had a human coworker and…sometimes it would make a mistake you’d never have expected...you’d have a pretty complicated relationship with that coworker. I still think [we’re] in that phase…” — Mike Krieger ([10:04])

• On Enterprise Blockers:

  • “A lot of the work that we are doing for next year is...distributability...If we want to bring our intelligence and even our agentic primitives...into actual enterprise workloads, we need to really embed and meet them where they are.” — Mike Krieger ([27:00])

• On the Vision for AI in 2026:

  • “Reliably take work off your plate.” — Mike Krieger ([30:01])

Key Timestamps

• Origins of Anthropic’s Coding Focus ([02:05] - [03:28])
• Rise of the Coding Agent (Claude Code & Hackathons) ([04:39] - [09:20])
• Non-technical and Tinkerer Adoption ([09:40] - [11:26])
• Rewiring Habits for Vibe Coding ([12:17] - [13:36])
• Three Buckets of AI Adoption ([17:23] - [21:25])
• Enterprise Trends and Process Redesign ([22:16] - [25:28])
• Technical and Organizational Blockers ([27:00] - [28:42])
• From Tool to Colleague ([28:42] - [29:50])
• 2026 Vision Statement ([30:01])

Conclusion

Mike Krieger and NLW paint a picture of an AI landscape on the verge of transformation: 2025 was the year coding agents hit mainstream, enabling both technical and non-technical users to partner with AI in new ways. Yet challenges remain, especially around reliability, integration, and reimagining workflows. Enterprises will spend 2026 focused on infrastructure and truly baking AI into their core products and processes. As AI agents become more capable and trustworthy, the promise for 2026 and beyond is clear and ambitious—AI that can finally “reliably take work off your plate.”