In partnership with
Gladly Connect Live '26. May 4–6 in Atlanta.
AI has everyone talking. Not everyone has answers. At Gladly Connect Live, CX leaders from Condé Nast, Smith Optics, and more share exactly how they moved AI from pilot to production, the timeline, the systems, the QA loops. 13+ sessions built for the moment we're all in. For CX and ecommerce leaders. Atlanta, May 4–6. Space is limited, secure your spot now.
Hey everyone!
I’m doing a quick roll call to see who’s hanging out in our community these days. I'd love to learn more about what you do, the industry you're in, and what you're most curious about right now.
If you've got 60 seconds, tap the link below and let me know!
Today:
OpenAI’s GPT-5.5 Pushes AI From Chatbot to Full Work Partner
Google Unveils New TPUs Built for the Agentic AI Era
xAI Launches Grok’s Fastest Voice Agent for Real-World Calls
ChatGPT Gets Team Agents That Can Work While You’re Away
Google Workspace Turns Gmail, Docs, Drive, and Slides Into One AI Brain
OpenAI has officially launched GPT-5.5, positioning it as their "smartest and most intuitive" model to date. Instead of just acting as a conversational chatbot, GPT-5.5 is designed to be a highly capable, autonomous digital worker. It is built from the ground up to handle messy, multi-step tasks across complex workflows—like coding, scientific research, and data analysis—with significantly less human guidance than previous generations.
The Important Details:
True "Agentic" Capabilities: You no longer need to hold the model's hand with step-by-step prompting. You can give GPT-5.5 a complex goal, and it will autonomously plan the steps, use tools, operate software, verify its own work, and navigate roadblocks until the job is done.
Speed Meets Intelligence: Usually, making a model smarter makes it noticeably slower. However, GPT-5.5 manages to match the blazing-fast latency of GPT-5.4 while operating at a much higher level of intelligence. It also uses significantly fewer tokens to accomplish the same tasks, making it highly efficient.
Topping the Benchmarks: The new model is already breaking records. It scored an 82.7% accuracy on Terminal-Bench 2.0 (which tests command-line workflows), narrowly beating out Anthropic's highly anticipated Claude Mythos Preview and reclaiming the top spot for publicly available models.
Pricing Shift: While efficiency is up, the raw API cost is increasing. Once available to developers, the GPT-5.5 API will be priced at double the cost of 5.4 ($5 per 1M input tokens / $30 per 1M output tokens). There is also an even heavier "GPT-5.5 Pro" tier for maximum power.
New Cyber and Bio Guardrails: Because the model is so capable at finding security vulnerabilities and parsing biological research, OpenAI is heavily restricting certain features. They introduced a "Trusted Access for Cyber" program so only verified security defenders can use its raw cyber capabilities. They also launched a $25,000 "Bio Bug Bounty" to challenge researchers to find jailbreaks in the model's biological safety guardrails.
Availability: GPT-5.5 is rolling out today! It is currently available in Codex and ChatGPT for Plus, Pro, Business, and Enterprise subscribers.
Google just announced its massive 8th-generation custom AI chips—the Tensor Processing Units (TPUs). But what makes this launch special is that Google isn't just focusing on raw power; they have explicitly designed these chips to handle the heavy, iterative demands of AI "agents" that reason, execute multi-step workflows, and swarm together to solve complex problems.
Two Specialized Chips: Instead of a one-size-fits-all approach, Google split the architecture into two distinct chips. The TPU 8t is built for massive, compute-intensive model training, while the TPU 8i is optimized for high-speed, latency-sensitive inference (perfect for deploying fast-acting agents).
Mind-Bending Scale: A single TPU 8t superpod can now scale to 9,600 chips and 2 petabytes of shared high-bandwidth memory. It delivers an unbelievable 121 ExaFlops of compute power.
Eliminating the "Waiting Room": When AI agents collaborate, even small delays can bottleneck the whole system. The TPU 8i was redesigned specifically to eliminate inefficiencies in inter-chip communication, allowing specialized agents to work seamlessly together in real-time loops.
Efficiency: The new system delivers nearly 3x the compute performance per pod over the previous generation, optimizing power usage while maximizing productive computing time.
We have all experienced the pain of talking to a robotic, confused automated voice system on the phone. xAI is trying to fix that forever with their new "Grok Voice Think Fast 1.0" model. Released as a tool for developers and businesses, this voice assistant is specifically designed to handle the messy, chaotic reality of human phone calls without missing a beat. It is already battle-tested and currently powering customer support and sales for Starlink.
The Important Details:
Built for the "Real World": This AI is designed to handle tough conditions like loud background noise, heavy accents, and people interrupting it mid-sentence. It natively supports more than 25 different languages, making it ready for global use.
Flawless Data Entry: Trying to read an email address, a physical street address, or an account number to an automated system is usually a nightmare. Grok Voice is built to seamlessly collect and repeat back difficult information—even if you speak quickly, stumble over your words, or naturally correct yourself halfway through.
Thinking Without Pausing: Usually, if an AI has to solve a complex problem, you have to wait in awkward silence for it to process. Grok performs its complex reasoning in the background, meaning there is zero added delay in the conversation. It responds instantly while still giving highly intelligent answers.
Harder to Fool: Many voice bots suffer from "hallucinations," meaning they will confidently give you a completely wrong answer just to keep the conversation moving. Grok Voice has built-in safeguards to quickly think through tricky, edge-case situations before it speaks, catching obvious mistakes that other systems confidently get wrong.
Beating the Competition: The new model took the absolute top spot on industry leaderboards that measure how well voice bots handle realistic, noisy phone calls. It beat out major competitors from Google (Gemini 3.1 Flash Live) and OpenAI (GPT Realtime 1.5) by a wide margin in specific scenarios like retail returns, complex airline booking changes, and telecom billing disputes.
🧠RESEARCH
Modern AI often relies too much on external tools (like calculators or search engines), even for simple tasks. Researchers found this happens because AI misjudges what it already knows and is rewarded for final answers rather than efficiency. By changing how models are trained, they cut unnecessary tool use by over 60%.
When AI helps with school or work, it can be hard to tell if the human actually understands the material. This paper proposes a new "governance framework," which is a set of rules for judging work. It allows AI use during the messy drafting stages but requires the final result to be understandable and usable without the AI's help.
AI agents often struggle with long coding projects because they forget previous attempts or get stuck in loops. Researchers developed a system that turns these long sessions into "structured summaries," highlighting only the most important parts. This helps the AI learn from its own mistakes and successfully finish difficult software engineering tasks more often.
📲SOCIAL MEDIA
🗞️MORE NEWS
OpenAI Workspace Agents OpenAI has launched new digital assistants for businesses that can independently handle complex, multi-step projects. Unlike regular chatbots, these programs run continuously in the background and keep working even when you close your computer. Company managers maintain strict safety controls to determine exactly what files and systems these automated workers are allowed to access.
Google Workspace Intelligence Google announced Workspace Intelligence, a new system that links together apps like Docs, Sheets, and Gmail to understand your projects and do the work for you. Instead of just answering questions, this tool can automatically organize messy data into interactive charts, create lifelike video characters for presentations, and write detailed meeting summaries. It learns your personal work style over time to help you get repetitive tasks done much faster.
OpenAI Privacy Filter OpenAI introduced a new tool called Privacy Filter that automatically finds and hides personal information inside large amounts of text. Because it is small enough to run directly on a user's computer rather than needing an internet connection to a large server, it keeps sensitive data much safer. This allows software creators to build stronger privacy protections into their own programs without the risk of data leaks.
Gemini Enterprise Agent Platform Google Cloud launched the Gemini Enterprise Agent Platform, a complete system for businesses to build and manage digital assistants that can perform complex, multi-step tasks on their own. The platform acts as a secure control center where companies can give these digital workers specific jobs while keeping a close eye on their actions and permissions. This helps businesses safely use artificial intelligence to run their operations without losing control over their private company data.
Anthropic Probes Unauthorized Access to Mythos The artificial intelligence company Anthropic is investigating how a group of unauthorized users broke into its highly restricted cybersecurity model, named Mythos. The intruders gained access through an outside partner company rather than breaching Anthropic's own internal security systems. Because Mythos is designed to find dangerous computer system weaknesses, authorities are deeply concerned about the risks of this powerful tool falling into the wrong hands.
Merck and Google Cloud AI Partnership The pharmaceutical giant Merck is partnering with Google Cloud in a massive deal worth up to one billion dollars to bring advanced artificial intelligence into every part of its business. Google experts will work side-by-side with Merck employees to build digital tools that help discover new medicines faster and manufacture them more efficiently. This massive investment highlights how major drug companies are racing to use new technology to speed up the delivery of new treatments to patients.
Anthropic's 81k Economics Study Anthropic surveyed over 81,000 users of its software to understand how artificial intelligence is impacting the economy and workplace morale. The study found that people who use these tools to drastically speed up their work are actually the most worried about losing their jobs in the future. Interestingly, workers in both the highest-paying and lowest-paying professions reported the biggest boosts in how much daily work they could accomplish.
AI Table Tennis Robot A new robot created by Sony has successfully defeated highly skilled human athletes in competitive table tennis matches. This milestone proves that computer programs can now process information fast enough to handle the lightning-quick physical movements and split-second decisions required in real-world sports. Experts consider this a massive leap toward building physical machines that can smoothly navigate and interact with our unpredictable world.
