Google is on the path to an AI-powered ecosystem of products. The company has been adding AI tools to its most used services, and the tools just keep on coming. According to a new report, Smart Reply on Gmail just got a boost thanks to Gemini.

Times truly have changed! Back in 2017, the height of AI was Google Assistant reading you a bedtime story. That was the year that Google introduced Smart Reply, a feature that would suggest replies to give people that you’re having a conversation with. The replies were pretty simple, and not much has changed since then. When you’re in an email conversation, you’ll see the smart replies, but they’ll still be pretty simple and broad in nature.

Smart Reply in Gmail will get the Gemini treatment

A feature like Smart Reply was not safe from Gemini, as it’s a feature that could very well be enhanced. Right now, you can only use it with the Gmail mobile app. As you can imagine, the enhanced Smart Reply will use context awareness. Basically, it will read the email conversation to gain context for what it’s about. Then, it will give you a selection of three different responses that you can send. Those will appear in a carousel at the bottom of the screen.

The suggestions will consist of two parts. The first part will be a line of blue text explaining the direction the response will go in. In the screenshots below, we see that one of them says “Confirm Sunday Launch, ask goals” and another one says “Confirm Sunday Launch, excited”.

So, for the first one, we’re sure that the response will confirm the launch and ask what the sender hopes to achieve with it. The second one confirms the launch and shows how excited the person is about it.

The second part shows you a preview of the text that will go into the text field. When you tap on the suggestion, it will go directly into the field so that you can review and edit it. You’ll want to look it over because generative AI is still relatively new, so there’s the chance that it could mess up on some details.

Who can use this feature?

Right now, you can only use this feature if you’re subscribed to Google One AI Premium or if you’re on the Gemini Business, Enterprise, Education, or Education Premium plan. So, if you’re a free user, you’re going to have to stick to the old-fashioned way of writing emails… which is writing them.

Spotify should be working normally again now after a few hours of service issues. The Spotify Status account posted on X Sunday afternoon to say, “Everything’s looking much better now!” Earlier in the day, it confirmed that it had been experiencing problems following users’ reports that the app and web player were acting up. Spotify users on social media reported a variety of issues, from songs repeatedly pausing on them to being locked out of the streaming platform entirely.

The problems spiked a little before 11AM ET, per Downdetector, and persisted for two hours or so. At 12:14PM ET, Spotify Status posted, “We’re aware of some issues right now and are checking them out!” In the meantime, the comments section of Downdetector turned into a full-blown unhinged group chat. I was still having issues with the web player around 1:15PM, but it all appears to have been resolved now.

Update, September 29 2024, 2:45PM ET: Spotify is back online. This story has been update to include a note from Spotify Status on the issue’s resolution.

Just as enterprises continue to adopt large language model-powered text-to-SQL as a way to ‘talk’ to their data assets, a new shift in the ecosystem has started emerging: AI agents. Today, New York-based Redbird announced a new chat platform that uses “specialist agents” to help enterprises handle most analytics value chain tasks, from data collection and engineering to data science and producing actual insights (reporting).

This means an enterprise user can give a natural language prompt to get insights from data in almost real-time and execute analytical efforts that pave the way for those insights. According to Erin Tavgac, the co-founder and CEO of the company, this represents more than 90% of an enterprise’s business intelligence efforts.

“For the past several decades the promise of truly self-serve analytics has fallen short for organizations, with the reality instead being complex data pipelines, dashboards, and shadow analytics that require technical skills. We have invested significant R&D into fusing the power of LLMs with Redbird’s robust end-to-end analytical toolkit in the form of AI agents that enable users to finally achieve self-serve, conversational BI that runs on their organization’s data,” he said in a statement.

Moving into the age of AI agents

While the age of AI agents is new, Redbird itself has been a long-standing player in the analytics domain. The company started in 2018 as Cube Analytics and provided enterprises with a no-code, drag-and-drop toolkit that enabled their users to create workflows aimed at automating and unifying all analytical tasks leading to dashboarding and insights. Earlier this year, the company expanded this work with the launch of a conversational interface, allowing users to ask business questions in natural language and receive insights and reporting outputs in real-time.

Now, as the next step, Redbird has added an ecosystem of specialized agents that operate on top of this end-to-end toolkit to orchestrate as well as execute multi-step analytical tasks to answer business-related questions.  

As Tavgac explained, admins setting up the chat platform have to choose a base LLM (like GPT, Llama etc) and load up their organization’s proprietary data ontologies, business logic and reporting blueprints (like business definitions, PowerPoint report templates, etc.) to customize it with relevant business context. Once the data is inputted, the AI agents using the LLM begin to use all the context and generate metadata from the information to do their work  — in response to user questions.

“User prompts are sent to Redbird routing agents, which identify the best specialist agents to execute the tasks for that prompt (like PowerPoint Reporting agent, Data Engineering agent, etc.) and figure out how to orchestrate the execution order of those agents. Each specialist agent then manages its own part of the overall task by identifying relevant datasets/ontologies and executing the needed task using the Redbird toolkit, which includes applications and functions to handle the mechanical steps of the pipeline,”  Tavgac noted.

Detailing the tasks, he noted Redbird agents can pull unstructured or structured data from over 100 data sources, including Snowflake, Databricks and Hubspot. It can run advanced processing on top of the collected data by performing data wrangling, AI-driven tagging and data science modeling. It can also generate robust reporting outputs (like presentations, Excel reports and email/Slack updates) while taking necessary actions based on those reports (like executing an ad buy/modifying a campaign).

“Once the task is executed, the chat platform responds to the user with not just a text answer but also any deliverables needed, like a PowerPoint report the agents built or the data that they collected from a SaaS system,” he said.

No-code workflow orchestration remains available

As enterprises double down on their data efforts, going beyond text-to-SQL — adopted by Dremio, Snowflake and many others – and streamlining the analytics pipeline end-to-end with AI agents could be a great way to save time and resources. 

However, as many may still have concerns over the reliability of AI agents, Redbird is not doing away with its original drag-and-drop interface for automating business intelligence workflows. Instead, the company has made no-code the secondary option for users. The agents will orchestrate the tasks while also creating a no-code version of the workflow, allowing users to audit and inspect everything in detail if required. 

“So far, existing AI solutions have primarily tackled the automation of a very small fraction of BI and analytics efforts (SQL querying). While Redbird values and solves for that use case (text-to-SQL), it is also applying the power of its AI agents to automate the other more difficult and more sizable parts of enterprise BI workflows… Our approach to solving this challenge has enabled us to onboard eight of the Fortune 50 brands and over 30 mid-to-large-sized enterprise customers in the last few months,” Tavgac added. This includes brands like Mondelez International, USA Today, Bobcat Company and Johnson & Johnson.

Currently, he said the company is offering its technology on a SaaS model with usage-based licensing fees and generating seven-figure revenue. However, he did not share the exact specifics.

As the next step, Redbird will continue its AI agent-driven work and take its new Chat platform to more enterprises. It also plans to add more advanced agents in the analytics value chain to enable even deeper AI-powered business intelligence coverage for non-technical users. 

“We also aim to expand beyond our primary focus on analytics / BI use cases and into a deeper ‘Large Action Model’ approach that leverages AI agents that can take more nuanced action based on the analytical results (i.e. purchase supplies, send invoices).

Kevin Ryan has had a long and storied career as a pivotal force of New York City tech. He’s the founder and CEO of investment firm AlleyCorp, which has invested in a wide variety of startups, and is a serial founder, participating in the early stages of companies such as Business Insider, Zola, Gilt, Pearl Health, and Transcend Therapeutics. He helped build ad tech company DoubleClick as president and CEO in the 1990s and early 2000s, and Google later bought it for $3.1 billion in 2007, transforming the online advertising industry. He went on to co-found unstructured database provider 10gen, which later changed its name to MongoDB and went public in 2017.

Last Tuesday, I interviewed Ryan to discuss pivotal moments in company transformation for the benefit of the companies chosen for this year’s Startup Battlefield 200 at TechCrunch Disrupt.

As a part of the Startup Battlefield 200 program, the selected founders participate in pitch training workshops as well as a series of exclusive master classes with top-tier VCs, successful founders and operational experts. The virtual program aims to prepare and excite them for what is to come when they exhibit, demo and pitch at Disrupt in October.

During Ryan’s session, he offered a lot of useful advice for companies at all stages, from finding a great cofounder, to when and how to seek funding, to how a founder’s focus should change as a company scales.

But given his background with DoubleClick and MongoDB, I asked him how company founders should decide when and whether to take an acquisition offer, versus when they should hold on and try to go public.

“There’s no formula but what I’m thinking about is, one, what do our prospects look like?” he said. “Let’s not be delusional — how much are we growing, what is this company going to look like in three years, what are the exit strategies, then how many other people — other buyers — are there, how are we doing relative to everyone else?”

He added, “Most people underestimate the time factor, so if we’re worth $100 today, four years from now it’s got to be worth $200 just to break even because of risk, cost of capital, things like that. So are you signing up as CEO [because you believe] that we’re going to be worth $300? If you really believe that then we should hold on. But if you just think it’s going to be $150 or $170 we should probably sell today because also you need to factor in: Markets can close at any time. You and I over 25 years could name many things we didn’t see coming. The Ukraine war. No one saw inflation coming. No one saw many things coming….and all of a sudden everything’s dead.”

By and large, he said, more people should sell earlier, rather than holding out to try and become the next Mark Zuckerberg, who famously turned down a chance to sell Facebook to Yahoo for $1 billion in 2006. (Disclosure: Yahoo owns TechCrunch.)

“I think more people should sell than probably sell on average,” Ryan told me. “You’re definitely going to read the story of the $20 billion company that turned something down, but there are a lot of other examples of people that could have [sold].”

He added that lot of founders don’t think clearly when it comes to personal wealth from an acquisition, chasing ever-bigger numbers instead of settling for a life-changing amount of money. And by not settling, they often end up with zero instead.

“I had this conversation the other day,” he said. “Someone could sell now and they’re going to make $30 million. $30 million is an incredible amount of money. It’s life changing, right? And they can… a year later go off and do so many things. And you know what? $60 million doesn’t make you much happier than 30, right, but 30 it makes a big difference from zero.”

He added, “It sounds great to make 60, 90, 100. It actually doesn’t change your life very much.”  

You can watch the whole interview here.