دانلود کتاب Agentic پیشرفته

Back in 1999, I got a phone call. My head of business development, Rudy, was in California and had an urgent message. “Hey, I met these folks here in San Jose, and they have something you ought to take a look at. It’s kind of a natural language-based command and control interface.” Thus began my agentic AI journey. And thus began my partnership with my fellow coauthor and dear friend, Babak Hodjat. What Babak had imagined was, in fact, pretty much what Rudy had described—how to operate devices and simple applications by way of textbased or voice-based commands. He had come up with the idea by challenging himself with “Wouldn’t it be cool to talk to my TV?” And he had it working. I flew across the Pacific from my then–home base in Hong Kong to meet with him and the team. Soon, I joined him and his startup, Dejima. The rest, including the DARPA project that led to the development of CALO (Cognitive Assistant that Learns and Organizes) and the Siri spin-off that was sold to Apple, as one says, is history. Fast-forward more than 25 years, and history is in the making again. For the first time on our great humankind journey, we can envision digital, artificial beings whose capabilities equal and surpass those of humans themselves, and not in narrow, special ized domains anymore, but on a holistic basis— across our vast field of knowledge, and, increasingly, across our vast field of actions. AI agents are a reality in 2026. How is that possible? What happened since the days Babak and I were busy, and having a hard time, convincing the world that a machine could act upon natural language commands? For a start, with the advent of ChatGPT in late 2022, natural language understanding became a solved problem. Gone were the days when one had to engineer ontologies to educate machines as to what a sentence actually meant. Through sheer brute force—essentially massive pattern recognizers, a lot of compute, and immense datasets—machines got to master language. No matter that no one had imagined this could happen so fast; the fact was that machines handled language as well as humans did, and in fact better than most humans did. Since most of humans’ actions are actually language-based (“Could you please make sure the updated budget is ready for next week’s board meeting and have finance send us the revised deck beforehand?”), it was only a matter of time until we figured the next logical step—plug large language models (LLMs) into machine command lines, or APIs (application programming interfaces) in order to start enabling language-based actions. So we need the updated budget for next week’s board meeting? Well, Claude Cowork, OpenAI’s Codex, or Google’s Antigravity will: 1. identify and source the existing budget files, 2. read from email, Slack, or Microsoft Teams exchanges, determine what the updates are about, and identify who owns them, 3. if need be, ping the relevant department heads and finance department primes for these updated numbers (or automatically compute them based on either the revised documents shared between team members or the latest CRM and ERP applicationssourced numbers), 4. plug in the revised figures into the master spreadsheet, 5. make sure the numbers actually make sense (are we within the CEO/CFO/board-mandated guidelines? How are we tracking against budget?, etc.), and 6. format that into a presentable deck, with a headline commentary “please,” for circulation and consumption a week prior to the board session. That is no longer magic. We are at an inflection point where any and every human-designed workflow is game for agentic AI enablement. The last few months have seen skyrocketing token consumption by LLMs, precisely because enterprises and individuals alike are discovering the incredible leverage that AI agents deliver. Babak and I wrote this book because we believe the agentic AI revolution will be immensely consequential for us all. Given the enormity of the stakes —machines taking actions on our behalf, machines having agency, humans gaining in productivity in unprecedented ways, workflow and knowledge workers being augmented then replaced by machines, humans having to ensure that machines behave in compliant ways—we think it is of paramount importance to explain what is happening, to help guide our fellow humans towards maximizing the beneficial use of agentic AI, and to outline the pitfalls and help mitigate them by establishing relevant guardrails.