David Ortiz

AI started as a party trick. Then it became a supercharged search bar. But the real unlock happens when you start treating it like your hardest-working employee. I recently sat down with Keith Ferrazzi to talk about this for Fortune—specifically, how we’re moving from AI that just "knows" things to agents that actually do the work. Like most companies, we have always had a "scale wall" in our way. We want to deliver the best possible service and engage every single lead, but you eventually run out of people and hours. Whether it’s in service or sales, the budget and headcount math eventually stops working. The breakthrough is moving beyond "better search" and into autonomous agents. It’s finally possible to treat every customer like your most important one—and every lead like your only one—without hitting those old constraints. Read the full piece here: https://lnkd.in/gdFjdhHh #AI #Agentforce #Salesforce

222

4 Comments

If you're building with Pipecat, this one is for you. We just launched the Tuner <> Pipecat SDK 𝘱𝘪𝘱 𝘪𝘯𝘴𝘵𝘢𝘭𝘭 𝘵𝘶𝘯𝘦𝘳-𝘱𝘪𝘱𝘦𝘤𝘢𝘵-𝘴𝘥𝘬 Now you can plug in observability and analytics in minutes and get real visibility into how your voice agent is performing in production, at scale. Because shipping the agent is only step one.  Understanding what’s happening in production is where the real work begins. Link in the comments 👇

108

12 Comments

Over the past week, more people are seeing in real time what many of us have been warning about for over a year: large, generic LLMs are breaking down when accuracy, trust, and accountability actually matter. The recent issues surfacing publicly around Salesforce and LLM behavior are not surprising. They’re symptoms of a deeper structural problem in the first-wave of AI mass adoption being rushed into production. The future of AI isn’t renting intelligence from the cloud. It’s owning it, understanding it, and trusting it. That’s the next wave of AI AskSLM is already building. Very few talk about what to do next because wrappers won't fix structural problems and cloud-only AI won't survive in regulated reality. The real shift isn’t bigger models or better prompts. It’s ownership, accuracy, and where AI actually runs. That’s the gap AskSLM has been engineering toward from the start, well ahead of the curve. #AIInfrastructure #EnterpriseAI #NextWaveAI #BeyondLLMs #AIBuilders #OnPremAI

5

AI pricing is in flux, and even the most established enterprises are iterating in public. Salesforce is on its third pricing iteration for Agentforce since late last year. This time, it’s positioning Agentforce as a clear value add that increases the value of their core offerings, while also driving adoption through its Agentforce add-on that includes unlimited usage. These changes reflect several challenges with pricing AI agents: - As usage grows and value increases, changing pricing too slowly leaves money on the table and erodes margins.  - The value provided must meet customer expectations, and pricing must reflect both.  - It’s difficult to forecast how pricing changes impact revenue and customer adoption. There is no “set it and forget it” pricing model anymore. Companies need to have the ability to evolve their pricing over time. To navigate these challenges, pricing strategy isn’t enough—you need infrastructure that can keep up. That means billing platforms, like Orb, that are built for experimentation. They support frequent pricing changes, model the impact of pricing changes before rollout, migrate customers seamlessly, and give real-time insight into usage and revenue. https://lnkd.in/gSc2HyC2

23

1 Comment

AI agents are useless if you can't manage them. Workday's new Agent System of Record proves that AI is shifting from "cool tool" to "managed infrastructure." At Zentric, we don’t just deploy bots: we build the operational layer that ensures they deliver results. How are you tracking your AI's ROI? #ZentricAi #AI #BusinessOps #Efficiency #FutureOfWork https://lnkd.in/gxfdA_jv

8

OpenAI just put $4 billion behind Forward Deployed Engineers. Google Cloud is expanding their FDE team. Cognition, Anthropic, NVIDIA, Box — all hiring. If you haven't heard of a Forward Deployed Engineer yet, clear your schedule. You're about to hear about nothing else. The FDE is becoming the defining role of the enterprise AI era — the person who sits between a frontier model and a Fortune 500 company and makes the whole thing actually work. Part engineer. Part consultant. Part therapist for executives who just learned what a token is. The demand is real. The urgency is real. The hiring processes companies are using to find these people? Completely broken. FDEs don't fit neatly into any existing hiring box. Too technical for standard GTM screens. Too client-facing for standard engineering loops. Too ambiguity-tolerant for any rubric designed in 2015 that nobody ever updated. So what happens? Interviewers calibrate differently. Candidates get inconsistent signals. Great people fall through the cracks because one interviewer weighted "can they code" and another weighted "can they handle a room." Works fine when you're hiring 10 FDEs. Doesn't work when you need hundreds. 150 FDEs from the Tomoro acquisition gets @OpenAI to launch. The infrastructure to evaluate the next 500 — consistently, with structured criteria, repeatable scoring, and decisions you can actually explain — doesn't exist yet. The model race is largely won. The deployment race is just starting. The teams that can hire FDE talent at scale will be the ones who close the gap between what AI can do and what enterprises can actually absorb. That's the problem Avrio is built for. Exciting time to be in hiring infrastructure. Slightly terrifying time to be running a manual ATS. #ForwardDeployedEngineer #EnterpriseAI #HiringInfrastructure #AI #nvidiainception

5

We failed to build AI to help Salesforce organizations. Our first multi-agent implementations looked perfect in demos but fell apart in real environments... classic. Our agents couldn't handle inconsistent data, got confused by undocumented workflows, and broke when they hit the reality of how Salesforce is actually used day-to-day. The problem wasn't the organizations- it was our approach. We were solving the wrong problem first, building deeply integrated advanced AI, when what was needed was a way to prepare existing systems for this evolution. So we went back to basics. We called it "AI Spring Cleaning", working alongside organizations to understand what makes Salesforce environments truly agent-ready. What we learned? Most of the work isn't AI-specific at all, and a lot isn't even Salesforce specific. It's just good software hygiene. To avoid asking teams to spend months manually documenting their Salesforce setup, we built AI agents that do the heavy lifting (because why not). - Automatically map your data relationships - Discover hidden workflow patterns - Identify integration bottlenecks Suddenly, the "spring cleaning" went from a 3-month project to a 1-week sprint. All that tribal knowledge was centralized in one place. After multiple iterations and hard-won lessons, we have a much clearer idea of what it takes to make sure Salesforce dev teams can get real value from AI. The question isn't whether you'll adopt it, but whether your organization is prepared when you do. Spring was about getting ready. Summer is about getting after it.

87

5 Comments

Software you ship is mostly software you didn't write. A typical production app pulls in hundreds of third-party packages at build time. Each one is a path into your environment, and most have never been reviewed at the depth your own code has. That's the part of the risk picture shifting fastest right now. Anthropic's Mythos and OpenAI's GPT-5.5-Cyber are the visible examples. Mythos surfaced bugs in OpenBSD, FFmpeg, and the Linux kernel that survived decades of human and fuzzer review. The operational story isn't the headlines. It's that reasoning models catch a different class of bug than fuzzers do. Fuzzers find what mutation can reach. Reasoning models find what requires understanding intent. The bugs in your dependency graph that survived a decade of fuzzing didn't survive because they were hard. They survived because nothing was reading the code with intent. That capability is now available to defenders. It's also available to everyone else. The exposure isn't in the popular packages. Those get scrutiny. It's in the transitive dependencies four levels deep that nobody has looked at since they were written. The FFmpeg flaw Mythos surfaced sat in a single line of code that fuzzers had executed 5 million times. Until recently, the cost of looking deeper was higher than the perceived risk. That math just changed. Anyone running reasoning models against their dependency graph in CI yet, or is everyone still in eval mode? The risk is no longer just the code you wrote. It's the code you pulled in, built on, and shipped with. #CyberSecurity #AppSec #SupplyChainSecurity

4

2 Comments

First Salesforce. Then MCP. Now n8n. The AlphaLoops node is live and verified on n8n. If your team runs on automations, you can now plug carrier data (DOT, contacts, fleet, safety, news) into any n8n workflow. 1,000+ apps on the other side of the wire. No code, no custom HTTP node, no integration tax. A few patterns we're already seeing: → Inbound lead enrichment before the SDR sees it → Slack alerts on target account safety events → Daily territory digests per rep → MC verification straight from the TMS The thesis: meet GTM teams where they already work. Salesforce for CRM-native teams. MCP for AI-native teams. n8n for automation-native teams. More surfaces coming. Link in the comments.

7

2 Comments

Salesforce's M&A playbook for dominance in the AI era: 39% of Salesforce's recent acquisitions came from companies they'd already invested in. This isn't traditional M&A. It's a multi-year predictive intelligence operation where Salesforce's 1,100+ portfolio companies are prepped and evaluated for acquisition. Now, Salesforce is applying this playbook to AI M&A. They're the most active acquirer of AI companies in 2025, the overall most active acquirer of AI agent startups, and their current corporate venture profile is filled with their next acquisitions. Their AI acquisition pipeline today: → 147 active AI investments → 37 AI agent companies backed → 46% of portfolio = top-decile startups based on CB Insights Mosaic score -- our measure of a company's probability for success → 31% showing above-average M&A probability They aren't guessing which technologies will win. They are funding strategic categories and running live experiments through their customer base. The playbook: ✓ Invest early (median 3.3 years before acquisition) ✓ Validate product-market fit through customer deployments ✓ Test integration points via partnerships ✓ Observe leadership under pressure ✓ Build technical due diligence Some acquirers buy blind. By the time Salesforce writes a check, they've already watched companies operate for 3+ years. While competitors debate "build vs. buy," Salesforce already has 147 AI companies in evaluation mode. In looking at the data, it's clear they're not catching up to the AI wave – they've been proactively positioning for it for years with their investment and M&A pipeline. Agentforce didn't appear overnight. The future of Salesforce AI is already being built, backed by 37 agent-specific investments, 147 AI bets, and years of systematic portfolio construction.

82

7 Comments

This is the real filter: if nobody is willing to put their name on it, it’s not automation — it’s noise. The AI layer has to be approval-native: policy checks, audit trail, and explicit handoffs to accountable owners. Otherwise “smart” outputs never become decisions.

1

Five years ago, the audience for developer documentation was 100% human. Today, nearly half is AI. As AI moves from experimentation to execution, the primary bottleneck for agents is no longer model intelligence - it’s context. Salesforce Ventures is proud to co-lead Mintlify’s $45M Series B alongside Andreessen Horowitz. In an AI-native world, documentation has evolved from a static resource into load-bearing infrastructure. Agents don’t operate on intuition; they operate on context. By building the knowledge layer for AI-native documentation, Han Wang, Hahnbee Lee, and the Mintlify team are ensuring that critical context is structured for the agentic era. It is rare to find a team acting on such a massive inflection point with this level of speed and precision. We are thrilled to partner with Mintlify as they build a vital piece of the next-generation enterprise stack. Read more on our thesis here: https://lnkd.in/eyUKKMQr

46

4 Comments

Everyone's talking about the next big AI model, but frankly, most companies are still years from effectively deploying the last one. We're facing an implementation chasm, where raw model power rarely translates directly into real workflow automation or tangible business ROI without deep custom engineering. The actual competitive edge now comes from mastering integration, fine-tuning for specific use cases, and operationalizing AI at scale. It’s about building the bespoke solutions that truly move the needle, not just chasing benchmarks.

2

A 100-year-old creamery doesn’t sound like the place you’d expect to find an “agentic enterprise.” #sponsored And yet. At Salesforce #TDX26, I sat down with Daniel Peter, CTO of Petaluma Creamery, to chat about a traditional business that’s running on a surprisingly modern playbook. Small team. Lean setup. Very deliberate choices about where AI actually earns its keep. Watch to see what actually happens when you take a traditional business and connect the right data to the right problems; then give a small team the tools to move fast on purpose. And there’s one moment in this conversation that most leaders should probably sit with for a second. If you’re navigating AI in a non-tech organization, you’ll want to see this. #SalesforcePartner

111

40 Comments

Waymo has unlocked a surprisingly specific use case in San Francisco: parents are trusting it to drive their kids to school. They wouldn’t put their kid in an Uber, but they’ll send them solo in a fully autonomous car. That was the moment it really clicked for me. If AI can handle real-time decisions where safety is on the line, it can handle a mortgage file. Document review, condition clearing, data extraction—it’s not even close in terms of complexity. The challenge isn’t whether it’s possible. It’s whether we’re designing systems that make better use of people’s time—because the manual work isn’t just inefficient. It’s holding the entire industry back. https://lnkd.in/gFVA9Yvk #MortgageTech #Automation #AIinLending

13

1 Comment

For most of the last twenty years, one of the smartest things you could build was a software company. But something is shifting. The barrier to building software is collapsing, and the real leverage may be moving to the people who know how to use these tools to fix real problems inside businesses. I just published my first ever Medium article why the SaaS dream that shaped the last decade may be fading, and why the next era might belong to the people who just figure things out. Would love your thoughts. https://lnkd.in/efSttmvP

43

6 Comments

10 years down at Salesforce. What a ride so far. Thank you to every customer, partner and legend who made it one. I’ve never been more excited to be working at Salesforce. 10 years, 10 things I've learnt and who from - Manage energy, not time. Carve more for what fills you up. Bank a reserve for the rough patches. — Paul Voges Founder mode is real. Bold, uncomfortable decisions move things fast. Marc Benioff is a generational talent. Be the founder/CEO of your own world. — Marc Benioff Wellness is the unlock for life, work and growth. Move your body. Reflect. Find your passion. Hard physical work is a microcosm of life. — Stuart Hamilton "in the cave you fear is the treasure you seek" - find the moments where fear, energy and action meet. Exposure and reps set you free. — Steve Morrow Org design matters more than people think. Build for autonomy, shared goals, fast decisions, and the language your customers actually use. — Matt Freedman Don't get tricked by busy work. The real work is at the coal face. Have the hard customer conversations. Pitch the business. Get on the tools. — Frank Fillmann Vulnerability & sharing builds trust fast. It opens pathways for others to stretch and grow. — Anne Bennett Back your team all the way, even when that means pointing them at opportunities outside your team or the company. It should be why you wake up. — JJ McConnachie Start. Just start. The antidote to anxiety is action. Action creates change. Change creates confidence. — Jeremy Smith Be nice and be consistent. Nice is not a weakness. It's the thing that unlocks talent who'll work harder for you than you'd ever ask. — Amy Weaver Let's go.

533

116 Comments

At Turing, we partnered with 𝐒𝐚𝐥𝐞𝐬𝐟𝐨𝐫𝐜𝐞 𝐀𝐈 𝐑𝐞𝐬𝐞𝐚𝐫𝐜𝐡 to evaluate how well frontier models like GPT-5, Gemini 2.5 Pro, and Claude Sonnet 4 handle Olympiad-level math reasoning. Using Salesforce’s Hard2Verify benchmark, our experts annotated 200+ long-form math solutions, spending 500+ hours applying a strict zero-tolerance evaluation rubric to assess step-level correctness and reasoning consistency. The result: 100% compliance with Salesforce’s standards and a dataset that reveals where today’s LLMs succeed, and where they struggle with deep reasoning. This case study highlights how human-led evaluation can guide the next generation of AI systems. Read the full case study here: https://bit.ly/43GWscx

108

4 Comments

The transition to GPT-5 powered agents for Salesforce has been fairly smooth this week. ~33% of deployed agents are now running on GPT-5. iDialogue will continue supporting GPT-4.1 until OpenAI makes any API deprecation announcements. *Really* excited about a couple announcements coming up around Dreamforce (Oct 12th). I'll share some previews here throughout September. Follow iDialogue for the latest news about an Agent-powered Salesforce.

4