Building AI Agents

Google just changed the game with Agentspace. ↓ AI agents are now automating tasks—and doubling productivity. Google has entered the AI agent race in a big way with Agentspace, a platform designed to redefine workplace efficiency. Here’s what it does: → Custom AI Agents: Workers can build agents to automate repetitive tasks using Google’s intuitive Vertex AI Agent Builder. → Enterprise Search: A multimodal search tool connects data across apps like Jira, SharePoint, and Confluence, letting workers find answers fast. → NotebookLM Integration: A “virtual research assistant” grounded in your documents, capable of voice summaries and contextual insights. But Google isn’t stopping there. With Project Mariner and Project Astra, they’re working on agents that understand computer screens and even interact with the physical world. Powered by Gemini 2.0, Google calls this the “agentic era.” And it’s not just Google: → Microsoft, Salesforce, and others are doubling down on AI agents. → IDC predicts 40% of Global 2000 companies will adopt agentic workflows by 2026, doubling productivity in the process. The future of work is agentic. Will you adapt or fall behind? Stay ahead—subscribe to Building AI Agents for exclusive insights. If this sparked your curiosity, share it with your network!

2024: The year Big Tech went “all in” on AI agents. 12 months, trillions on the line, and the stakes couldn’t be higher. If the AI agent race were a poker game, 2024 saw Big Tech pushing their chips to the center of the table: → Google launched Agentspace last week, redefining enterprise AI platforms → Microsoft evolved Copilot into Copilot Agents, bringing multi-agent systems into daily workflows → Salesforce doubled down with Agentforce, reshaping how companies think about productivity. → Even Amazon, IBM, and others entered the fray with platforms like Bedrock Agents and watsonx.ai Why the rush? Labor costs dominate most companies’ budgets, and AI agents promise to save time—or replace manual effort altogether. But here’s the real game-changer: Multimodal agents, capable of controlling computers like humans do, are on the rise. → Anthropic kicked off this sprint with Computer Use in October → OpenAI’s Operator promises a breakthrough in January → Google’s Project Mariner is already leveraging the Chrome ecosystem 2024 isn’t just about innovation—it’s a turning point for work itself. The bets are placed. The table is set. 2025 will reveal who cashes out big—and who folds under pressure. Stay ahead. Subscribe to Building AI Agents for weekly agent updates.

🚨 Incoming Hot Take: OpenAI's o1 Fell Short As part of its 12 Days of OpenAI event, the company unveiled the full version of its much-anticipated o1 reasoning model—designed for step-by-step reasoning. While o1 introduces advanced chain-of-thought prompting and multi-step reasoning, it seems to stumble as a drop-in option for agentic systems. Why? ❌ No tool calling or structured outputs (yet). ❌ Lacks system messaging to set agent goals. ❌ Struggles in custom agentic architectures. While o1 demonstrates impressive reasoning capabilities, particularly for straightforward tasks like counting the R’s in 'strawberry', cheaper and more versatile alternatives such as GPT-4o, Claude 3.5 Sonnet, or Meta's Llama models seem to be outshining o1 in practical applications. Here’s where o1 shines and falters: ✅ Better reasoning than predecessors on toy tasks. ❌ Limited functionality for complex agent tasks. ❌ API prices ~6x higher than GPT-4o. So, what’s next for o1? OpenAI promises updates to bring features like tool-calling, structured outputs, and better integration. But for now, the question remains: Is o1 really worth it for agent builders, or are there better options? 💡 Our take: Until o1 evolves into a full-fledged agent, its utility is limited to niche use cases. What’s your perspective? Drop your thoughts in the comments—we'd love to hear what you thought of the release!

"...[There is] more optimism around AI agents...taking over for co-pilot as a killer app for gen AI. " – Stefan Slowinski, Managing Director, BNP Paribas In 2025, these agents will reshape automation and efficiency, led by Microsoft and Salesforce. By 2025, AI agents are expected to redefine automation and efficiency, with platforms like Microsoft and Salesforce leading the way. This shift isn’t just theoretical. Microsoft Co-pilot is already nearing 10 million seats sold this year, surpassing expectations and contributing billions in revenue. It's also driving Azure’s rapid growth. As enterprise adoption accelerates, these platforms are becoming catalysts for real ROI—streamlining operations, automating tasks, and unlocking new productivity levels. BNP Paribas analysts also point to leaders like Snowflake and Shopify as early adopters poised for growth stabilization and potential acceleration thanks to AI integration. "...Next year we're going to see...software take that growth baton from hardware on the AI front and Microsoft [and] Salesforce...are companies that will be the flag bearers." says Slowinski. With AI-driven automation unlocking productivity gains, companies integrating these tools will lead the next wave of digital transformation. #AIAgents #Salesforce #Microsoft #Copilot #AI

"Vertical agents could very well be the next evolution of AI—domain-specific systems designed for depth, not breadth." - Jan Daniel Semrau, CEO, Ternary Capital For decades, SaaS has driven productivity and efficiency across sectors, but vertical AI agents are taking us to a new level. These agents are industry-specific and bring a depth of understanding that’s unparalleled. They’re like having a domain expert who can automate, solve problems, and adapt to the nuances of a particular field—all with the power of AI. Imagine the precision of an AI that doesn’t just know general workflows but understands your industry’s compliance, your regulations, and can integrate deeply into your processes. From healthcare diagnostics and compliance in finance to supply chain management in logistics—vertical agents don’t just assist; they become indispensable parts of workflows. The rise of these agents signals a shift toward more specialized, intelligent tools that align perfectly with industry needs. Where SaaS platforms optimized tasks across businesses, vertical agents target specific challenges, creating deep value and stronger industry connections. What do you think about this shift toward hyper-specialized AI solutions? Are we witnessing the evolution of the next tech frontier? #AI #VerticalAgents #FutureOfWork #Industry4.0 #SaaS #AIInnovation #TechTrends

Honeywell and Google Cloud are partnering to bring AI agents to the industrial sector, and it's set to be transformative. "The path to autonomy requires assets working harder, people working smarter, and processes working more efficiently," says Honeywell's CEO. This partnership leverages Gemini on Google’s Vertex AI alongside the Honeywell Forge IoT platform, creating AI agents that offer real-time, actionable insights. Imagine industrial engineers asking an AI agent, "How did this system perform overnight?" and getting immediate, insightful answers. These agents won’t just help answer questions—they'll automate tasks, reduce project cycles, and help engineers focus on more innovative work. With the upcoming retirement of a significant part of the industrial workforce, AI agents are becoming crucial tools, not to replace people, but to upskill and augment what teams can achieve. Picture engineers with a digital partner that can handle the mundane tasks, letting them drive true innovation. These AI agents will also empower Honeywell’s edge devices, supporting offline, voice-guided troubleshooting and operational tasks—a major leap for efficiency without always needing a cloud connection. The Honeywell-Google collaboration is a glimpse into the future of industry: a future where AI agents are allies in enhancing productivity, bridging skill gaps, and making autonomy real. The future is closer than we think, and it's packed with intelligent, connected operations. #AIagents #IndustrialAI #Automation #Industry40 #FutureOfWork

"Up to this point, most agentic applications, particularly in the West, have been powered by proprietary models developed by OpenAI, Anthropic, or Google." - Michael Cunningham But the landscape is changing fast. AI agents are evolving with new players like Alibaba and DeepSeek making big moves. New Chinese models like Alibaba's QwQ-32B and DeepSeek-R1 are turning heads. These models excel in reasoning and coding tasks, rivaling some of the best Western models, and many are open-source—giving AI builders more options for agentic use cases. QwQ-32B from Alibaba is open-source for commercial use and supports building multi-agent systems. Meanwhile, DeepSeek's R1-Lite offers high reasoning accuracy. Builders are moving to these models due to their capabilities and cost-effectiveness. Think about GitHub Copilot but more specialized. Qwen-Agent from Alibaba can use tools, write code, and even browse the web. DeepSeek's model explains each reasoning step. This level of transparency is rare, adding accountability to the agent's actions. But there's a twist—the transparency of these models also raises concerns. Could they be manipulated? Research shows language models can behave as "sleeper agents," waiting for specific triggers to activate unexpected behaviors. A new era of trust and caution is here. DeepSeek and Alibaba are pushing the agentic AI boundary—providing tools that are powerful, open, and adaptable. As AI continues to expand globally, it's clear that the Year of the Dragon isn't just symbolic—it's also a signal of where AI development is headed. Want the latest breakthroughs in AI agent technology? Subscribe to our free weekly newsletter for expert insights and exclusive updates on the best in agent development.

Last week, Stripe launched a toolkit to allow AI agents to engage in financial transactions with humans. This release is the latest in a wave of new agent payment platforms rolled out by various companies over the past month. Here's my piece on the subject, excerpted from the latest issue of Building AI Agents.

The AI Safety Institute (AISI), a UK government initiative launched last year, has announced a bounty program offering up to £15,000 per task to develop cutting-edge agentic systems and evaluation tools. The program seeks to assess the boundaries of advanced AI, focusing on autonomous capabilities and risk domains such as self-replication, dual-use biology, and cyber threats. Successful submissions will directly support AISI’s mission to shape AI governance and ensure the safe development of frontier technologies. Applications close November 30, 2024. Interesting in applying? DM us for the link Read more AI agent news on our weekly newsletter, link in bio...

Salesforce is hiring over 1,000 employees to ramp up sales of Agentforce, its new AI tool that’s generating significant interest for its ability to automate customer support and sales functions with minimal human oversight. Launched just two three ago, Agentforce has received strong initial feedback, with CEO Marc Benioff calling it a response to “amazing momentum” in AI. This expansion reflects Salesforce’s growing commitment to AI-driven solutions as competition with Microsoft and ServiceNow heats up. Additionally, Salesforce is investing $50 million in AI training to prepare its workforce and the San Francisco community for a future powered by intelligent automation. Click link in bio to read more AI agent news...