Two Weeks with Claude Opus 4.5: A Review

The author spent about two weeks using Claude Opus 4.5 firsthand and recorded their impressions along with the massive changes this model is poised to bring. Opus 4.5 goes beyond being a mere coding model -- it is the decisive "unlock" that enables agents to perform real-world work, and it delivers powerful synergy when combined with the Claude Agent SDK. This article covers everything from the mindset for embracing the agent era to concrete, practical tips for Claude Code and SDK usage.

1. The Big Picture: The Dawn of the Agent Era

Unlocking Agent Potential

Anyone who has used Opus 4.5 can feel that the pace of AI advancement has not slowed one bit. The author is surprised that people are not treating this moment as an even bigger deal. Perhaps the discussion was delayed because it launched right before Thanksgiving and overlapped with the NeurIPS conference, but Opus 4.5 is unquestionably the best model for coding and agent work.

If GPT-4 opened the era of chat and Sonnet 3.5 opened the era of coding, Opus 4.5 is the key that opens the era of agents. Agents can now reliably operate over increasingly longer time horizons and handle real-world tasks on our behalf.

It is like Waymo (self-driving cars). You say "take me from A to B" and it just takes you there. After a few experiences like this, your brain realizes: "Oh, I see. This is the world we live in now." From that moment on, you are completely captivated, and you can never go back to the old way of working.

2025, the Year of Agents

2025 has been called "the year of agents," and the combination of Opus 4.5 and the Claude Agent SDK has made that a reality.

What matters here is the "harness" (i.e., the SDK) just as much as the model. No matter how good a model is, it is useless if the harness is poor. The Claude Agent SDK provides a world-class harness for agents, and combining it with Opus 4.5 lets you build agents that actually work. The author borrows Alan Kay's famous quote to describe the situation:

Alan Kay said, "People who are really serious about software should make their own hardware." The agent version of this is "People who are really serious about models should make their own harness." Anthropic clearly believes this, and the results are magical.

There is enormous economic value hidden in this combination that we have not yet discovered, and it is now up to builders to extract it. If you have been skeptical about agents, now is the time to change your mind.

AI Made Real and Anthropic's Future

The author quotes Ilya Sutskever's interview, marveling at the mere fact that Opus 4.5 exists:

"You know what's crazy? All of this is real."

Watching Opus 4.5 work, the author found themselves leaning back and laughing, thinking "Wow, I live in a world where this exists." Current agent technology is that impressive.

Anthropic's growth trajectory is staggering. Revenue has grown 10x every year for three consecutive years (2023: $100M, 2024: $1B, 2025: projected $10B). CEO Dario Amodei was not sure whether this trend would continue in 2026, but the author believes Anthropic will exceed expectations. They even predict Anthropic could surpass OpenAI's valuation by early 2027, thanks to the best model in Opus 4.5, the best harness in the SDK, and the right strategy of focusing on the enterprise market.

2. Hands-On Review and Practical Tips

Now for a more practical discussion. Here are the strengths and useful tips the author discovered during two weeks of use. (For more firsthand accounts, check out this article.)

General Usage

Treat it as a colleague: Think of Opus 4.5 not as a simple tool but as a genuine collaborator. Trust it more than you think you should and delegate work to it. This helps rewire your brain for future workflows and pays off in the long run.
Assign more complex tasks: Trust the model and give it longer, more complex assignments. Reduce oversight, and if you never have to rein it in for being too aggressive, you probably are not trusting it enough.
Ramble via voice: If you are not using voice input yet, you are living in the Stone Age. Just pour out your thoughts by speaking for five minutes, and Opus 4.5 will organize it into polished output like a great team member.
Efficiency and image recognition: Opus 4.5 is more efficient than Sonnet 4.5, and image input has been significantly improved. Screenshot-to-code, in particular, is on an entirely new level.
Obsidian and Computer Use: Try using it with Obsidian (related YouTube video) or test the Computer Use tool. It is not perfect yet, but this feature will be mainstream by 2026, so get familiar with it now.
Agent Swarms: Try the "swarm" approach where multiple agents collaborate. Create a chatroom.md file for agents to communicate with each other, or leverage the SDK's Hooks feature for creative workflows.

Claude Code

The world's best coding tool: The Claude Code + Opus 4.5 combination is indistinguishable from magic. Whether you use it in the terminal or the desktop version, this is the current pinnacle of coding AI.
Plan Mode: The new Plan Mode is remarkable. Not using it for complex tasks is leaving significant performance on the table.
Compacting issues resolved: Previously, performance degraded when conversations got long and content was compressed. No need to worry anymore -- Anthropic has solved this brilliantly, and Opus 4.5 can infer omitted context with uncanny accuracy.
Design capabilities: Leverage frontend design skills and screenshot features to produce decent designs. AI-generated interfaces have now reached a "good enough" level.
Best-of-N: In the real world, you cannot have five developers build the same thing and pick the best result -- but with AI you can. Opus 4.5 excels at proposing multiple approaches, explaining trade-offs, and guiding you to the best choice.
Write pseudocode: Sometimes, instead of typing in a chat window, write comments or pseudocode directly in the code file describing what you want. Opus 4.5 will understand your intent perfectly and implement it in code.

Claude Agent SDK

The best open secret: As mentioned, the Claude Agent SDK is the open secret weapon of the AI industry. Combined with Opus 4.5, it becomes unbelievably powerful.
Go deep: Do not skim the surface. The more you know about the SDK, the more you can accomplish. Agent proficiency is a real skill.
Build your own agents: As a weekend project, try building an agent that automates three things you frequently do on your computer. Once you succeed, you will become addicted to automating more -- and this is a very productive addiction.
Cloud deployment: For beginners, deploying agents to the cloud can be somewhat challenging since you need a server that supports long-running agents in a sandbox environment. Services like E2B offer a good developer experience (DX), so push through the initial learning curve. You can have an army of agents working while you sleep.

3. Closing Thoughts

Claude Opus 4.5 and the Agent SDK are ready to fundamentally transform the way we work. This is not merely a tool upgrade -- it is the unlocking of an entirely new dimension of work.

Now is the time to ride this wave. Trust the agents, study them deeply, build with them, and make 2025 a truly transformative year. Anthropic will continue to win, and those of us who leverage these tools well will win alongside them.