@sukh_saroy Anyone that has done serious work with Agents know that they are a long way of doing production-grade work without good orchestration and/or supervision. Doesn’t matter how many slop comments and slop article reviews
Claude Code's capabilities are incredible. However, one area that still needs a lot of development is the use of APIs and environment variables. Perhaps due to the need for security and sandboxing, it often gets lost, and a simple API call sometimes takes dozens of prompt adjustments
While many people focus on big accomplishments, sometimes it's the little things that truly matter. I have been using Cloud Cowork lately for building slide decks, and I was amazed at how well these tools now follow guidelines, branding, and templates. This was not possible just a few months ago. The productivity gains from small, repetitive activities like creating slide decks, updating spreadsheets, or doing online research for prices and budgets are significant. Though each of these actions may seem minor, their combined effect greatly increases productivity for individuals and organizations alike.
@emollick The interesting bit is that it’s all pretty much trial and error now. We are still early days on how to build tools that are somewhat “scientific”. It's amazing how much we can achieve with the current models alone in terms of productivity
@theo I was thinking about the exact same thing this weekend. Great article! There is also the multiple agent problem (codex in one tab, CC in the other)…
The evolution of programming? With agentic development, there may be a time when we won’t need keyboards or mouse’s anymore, we can just chat with the computer. Do you think that will happen someday?
Something that Claude Code and other Code Agents still need to improve is following strict guidelines without overly complicated setups. That is specially true for CI/CD and common git practices. Even after multiple instructions, if you are not careful Claude will commit to main.
The recent hype over Clawdbot (an open-source Agent that can reliably handle day-to-day tasks like organizing emails, sending messages on WhatsApp, etc.) may be an early indicator that 2026 could be the year Agents reach the mainstream B2C consumer.
Recent moves like Apple bringing Gemini capabilities to Siri and Anthropic launching Claude Cowork go in the same direction, moving away from Chatbots toward agents that can interact and operate devices on the user's behalf.
For most people, this should be the year that smart assistants finally become smart. For companies, there will be an immediate need to make products AI-ready so agents can interact reliably and safely.
There is obviously a significant cybersecurity risk that must be addressed. Companies that have deferred their IT and Data estate modernization will also have to move fast or risk becoming obsolete.
Claude in Excel is now available on Pro plans.
Claude now accepts multiple files via drag and drop, avoids overwriting your existing cells, and handles longer sessions with auto compaction.
Get started: https://t.co/cAMDXM1h7r
@GergelyOrosz Did high level languages replaced Assembly programmers? not entirely, but the share of Assembly programmers reduced a lot. I guess something similar could happen, replacing "hard core" soft eng by a new PM/SE role.
This is a behavior we will start to see more and more as agentic development become mainstream. To avoid unexpected behavior (or risky behavior), tools will implement explicit consent gates that require human intervenction.
The image below is from the Prisma ORM, a package that manages interaction between a server and a database.
It’s crazy how Claude Code improved the agentic development process compared to just 6 months ago… it’s fair to say that development is already changed for good. There is no turning back
They call it getting “Claude-pilled.”
It’s the moment software engineers, executives and investors turn their work over to Anthropic’s Claude AI—and then witness a thinking machine of shocking capability, even in an age awash in powerful AI tools. https://t.co/sm2yyLTsev
There is a real challenge in implementing AI-assisted data engineering (what I'm calling "Agentic Development"), which is how to get the validation process right. In software engineering, we have tools that guarantee the code works technically before it runs.
Data is messier. You can have perfect code, but if the data input is bad, the output is bad. Validation here means checking both the logic and the actual numbers.