logo
AI agents are an opportunity to rethink creativity: Adobe's Govind Balakrishnan

AI agents are an opportunity to rethink creativity: Adobe's Govind Balakrishnan

Hindustan Times24-04-2025

While the generative artificial intelligence (AI) battles continue in earnest, headlined by the likes of OpenAI, Google and others with the definitive leaps they make with new models, there is one focused AI product that continues to deliver within a focused set of apps. Adobe's Firefly models, which can be further segregated depending on images, vectors or videos, have clocked more than 22 billion asset generations. The new Image Model 4 and Image Model 4 Ultra will underline Adobe's AI efforts across apps including Express and Photoshop. At this year's Adobe Max London conference, the company has detailed plans for general availability of Firefly Video models addition of non-Adobe models (GPT image generation, Google Imagen 3 and Veo 2, Flux) as choice for creators, an upcoming Firefly app for smartphones, and significant upgrades across Photoshop, Illustrator, Lightroom, Premiere Pro and InDesign.
The versatile Adobe Express platform is adding significant new functionality too, as it competes with the likes of Canva as well as a number of AI-based tools that have become proficient at replicating some of Express' functionality. The new additions add video generation, enhancing speech to remove background noise, animation for objects in static designs and an AI clip maker that creates shorter duration clips from a larger video, optimised for social media sharing. Adobe is taking forward steps with AI Agents, across Photoshop and Premiere Pro.
Also Read:Agentic AI: Next big leap in workplace automation
On the sidelines, Govind Balakrishnan, senior vice president for Express Product Group & Digital Media Services at Adobe, spoke with HT from London, and detailed India's importance as a market for Adobe Express, how important it is to leverage Firefly amidst stiffer competition, if Adobe Express' perceived value proposition has changed with time and the excitement around AI Agents (Adobe has detailed plans to bring agentic AI to Express). Edited excerpts.
Adobe Express has come a long way since the big revamp in 2021. Did Adobe, at that time, envision such a thick AI layer to envelope Express over time, and is Adobe Express today exactly as you'd thought of it then?
The benefit of starting with a complete free platform is that we have an ability to think holistically about where we think the product should go, and we obviously have the advantage that we were beginning to see what was happening in the industry around us, including advancements that were happening in AI and more. Importantly, even the advancements happening in generative AI. One of the key tenets that we wanted to hold true to was that we would provide recommendations, that is we would not put the burden entirely on users to figure out what to do and when to do it. From the beginning, we architected the product in a way that we could provide font recommendations. We give our users access to more than 30, 000 fonts now, but that doesn't help if you can't figure out how to pick the right one. Look at the content, context, and we make some recommendations. Similarly, we are also making colour, stock assets and other recommendations. Users don't have to go searching for these pieces of content.
Then along came generative AI. We were still in the middle of building the product when generative AI started taking off as much as it did. Adobe has deep experience in generative AI base, and how we got Firefly to where it is today. But since we were building the product, we didn't have to bolt it all as an afterthought. We have actually seamlessly and contextually integrated it with the workflows in the journey. While you're creating, just add something as simple as video generation, that we just did. Now, it's not something you start with. You actually start with your intent. You can start with video generation if you choose to, but the more natural way is, you're now creating a Instagram reel for instance and as part of that, you're putting some clips together. Now, you need a five second clip that is a filler potentially? So, in that workflow, in that context, we give you the ability to generate video. It's that integration seamless integration as part of your workplace that has been a big advantage for us.
Also Read: Video Gen AI battles begin, as Adobe releases Firefly Video model into the world
Did we expect the product to be what it is today? I would say, not necessarily. I take pride in is the fact that the product has evolved based on what we are seeing and how we're seeing our customers use the product. So, rather than us sort of saying, you know, this is exactly what the product should be, and this is how all the workflows and experiences should be, we have chosen to morph, adjust and build the product based on what we hear from our community and our users
How pivotal has Firefly been to Express, as it has been to the likes of Photoshop and Premiere Pro for instance, and does that need to be Adobe's AI trump card considering the AI ecosystem is creating focused alternative after alternative at a rapid pace?
I would say yes, and for a few reasons. We have essentially created the creative category, and we have more experience and expertise in the creative space than any other company in the world. Our researchers have been hard at work for many years, making sure they build the models that are best in class. There will be pointed solutions that will come and obviously challenge us, which is great. But if you look at it holistically, I would say that our ability to deliver the best results is real and is something that we are investing in and stay ahead of everyone else.
What then differentiates our Firefly models is that what we have built is safe for commercial use. We have trained these models with data that we have access to and essentially what we have done is, regardless of whether you're a student, a solo planner or someone in a business, we have given you the ability to leverage generative AI capabilities and use it for your creative work without being concerned about your IP infringement of any kind. That's the other one that we have held through to.
Also Read: We aren't building AI models for the sake of it: Adobe's Deepa Subramaniam
The third piece to this is that because we have built the tools that work on content. We have been able to bridge those solutions together and make those available through these seamless, contextually relevant touch points. The integration works better unlike a point solution that can be used to generate a stunning image. The questions, what do you do with that image? The image is not an end to itself, and you generally have to do something with the image. Perhaps put it into a poster, into a birthday card, or a social media campaign. We can now make all of that available to one single tool that gets you to start from an intent, use generative AI wherever your design, and then take it out with the flexibility across social media to banners to flyers. Users now have an end-to-end solution that only Adobe can provide right now. It's a matter of plugging these pieces together and obviously continuing to make sure we deliver the highest quality delivered safely.
Do the contours of perceived value proposition for Adobe Express change with time, and surely that ties in with what the consumer and business audiences are looking for?
For consumers, in general, their primary goal is to get started with an intent as soon as possible. We give them the ability to either start with a template, their own content, or generate their own images or videos. You can essentially enter a prompt to generate a design, modify it and get to desired outcomes. As part of that journey, users are able to use our generative AI in a meaningful way. When I say generated, it relates to image generation, video generation, text effects and an ability to create captions for social content.
With Enterprises, we are seeing a significant traction, and adoption, for generative AI in particular. That is given the quality of what's created or generated, but more interestingly, also because whatever we generate for commercial use is proving to be something that a large number of enterprises care about and value. The fact that it's safe for commercial use, and it's the confidence that there is no IP infringement. Those are the two things that go hand in hand as it plays to the value proposition that we are offering with generative AI to enterprise customers.
Also Read: Our AI innovation undergoes careful evaluation and diligence: Adobe's Grace Yee
AI agents are now very much part of the conversation for businesses, and perhaps in due course, for consumers too. What is Adobe's vision for agentic AI in Adobe Express and what shall form the core of this evolution?
This is an area once again that I get very excited about and often tell my team in the context of the advancements that we are seeing, and the tools that we have in our toolbox, are AI agents, now really a path to, or a way to interact with these applications. The opportunity that we have is to essentially reimagine and rethink creativity. This next wave, with the capabilities that are available and with the direction that we are headed in, we're seeing just the tip of where we want to go. As we deliver on that vision, we think a user will now have the ability to essentially interact with the application, purely using prompt based interaction models. It can be conversational or input text based and a user will no longer have to take the time and energy to essentially learn a tool like Adobe Express to get work done.
Think of it as the ability to bring what's in your mind's eye, to a digital service. No need to learn a tool to do that. What makes us even more magical though is that even as a user goes through that process, is the granular control to get to your desired outcome. The tools that we have in a product like Adobe Express are still available and a user can very seamlessly go from interacting using prompts to actually going into the tool and making the tweaks and adjustments manually. Users are not giving up on control by adopting, but enhancing an ability to start using these interfaces and capabilities to then get to the desired outcome as well as possible.
Also Read: Video Gen AI battles begin, as Adobe releases Firefly Video model into the world
Please tell us about India as a market for Adobe Express. What are users here demanding, and does that feedback and research help build for the rest of the world?
India is one of the most unique markets in the world, through diversity and its dynamic nature. I strongly believe that it is one of our biggest growth areas as we look ahead, not just for Adobe Express, but more holistically for Adobe. Given how creative the Indian population is, India presents a huge opportunity. One of our biggest challenges previously was that we did not have the tools and applications that could lower the barrier to entry. To address the broad base of users in India, Adobe Express is bringing all these capabilities together and we believe we have lowered the barrier to entry sufficiently enough for us to have lots of viable and exciting offerings for India.
We are going to invest heavily, and even based on investments and the advancements we have made, we are seeing a 3x increase in the number of Adobe Express users in India over the past 12 months. We expect that number to continue to grow. Our focus is on investing in making sure that we have the right content for the Indian market, be it templates, stock images, or forms. We are doing everything we can to increase the breadth of content that we have available for users in India.
In parallel, we are also trying to build solutions such as an AI assistant and agent solutions that don't necessarily rely on templates and content — a user simply explains what they want to do, and generate the right content. We will also add support for additional languages. We recently added a few additional languages in terms of localisation of the UI itself, but beyond that, we are making it so that users can translate content into as many as 15 Indian languages.
Adobe is also partnering with educational institutions and the education industry at large, the Ministry of Education and we have a number of compelling partnerships in the works. There is a big opportunity in education for kids and students in schools and educational institutions to use Express for their learning and creative work. If we put it all together, it is a big opportunity for us in India. It is a priority and we intend to stay focused.

Orange background

Try Our AI Features

Explore what Daily8 AI can do for you:

Comments

No comments yet...

Related Articles

WWDC25: Apple keynote at 10:30pm on June 9 - where to watch, what to expect
WWDC25: Apple keynote at 10:30pm on June 9 - where to watch, what to expect

Business Standard

timean hour ago

  • Business Standard

WWDC25: Apple keynote at 10:30pm on June 9 - where to watch, what to expect

Apple kicks off its annual Worldwide Developers Conference (WWDC25) today, June 9, with a keynote event scheduled for 10 am PT (10:30 pm IST). The conference will run through June 13, during which Apple will unveil what's next across its software platforms. The keynote will be live-streamed on Apple's website, official YouTube channel, and the Apple TV app. You can also watch it live through the video embedded at the end of this article. Apple WWDC25 keynote: Event details Venue: Apple Park, Cupertino, California Time: 10 AM PT (10:30 PM IST) Livestream: Apple's website, official YouTube channel, Apple TV app Apple WWDC 2025: Event schedule Following the keynote, Apple will host the Platforms State of the Union session, offering a deeper look at the new tools and technologies available for developers. This session will be available to stream via the Apple Developer app, the Apple Developer website, and Apple Developer's YouTube channel. On-demand playback will also be made available. Throughout the week, Apple will host over 100 technical sessions to provide insights into the company's latest frameworks and software capabilities. These will be accessible through Apple Developer platforms, along with supporting documentation and developer guides. Apple WWDC 2025: What to expect Platform updates Apple is expected to unveil a major visual overhaul across iOS, macOS, iPadOS, watchOS, tvOS, and visionOS. A new unified design language may include translucent menus, motion effects, and consistent interface elements across devices. In addition, Apple may introduce a new versioning scheme for its operating systems. For instance, the upcoming iPhone software could be branded as iOS 26 instead of iOS 19, bringing naming consistency across the ecosystem. Beyond aesthetics, Apple is likely to announce updates to its native apps. Notably, Game Centre is expected to become a cross-platform gaming hub for launching games, tracking achievements, and accessing curated content. The Health app may also see AI-powered wellness and tracking features added. Apple Intelligence and Siri While not expected to take centre stage, Apple Intelligence may still receive meaningful upgrades. A new software development kit (SDK) could allow developers to better integrate Apple Intelligence features like Writing Tools into their own apps. Apple may also deepen partnerships with third-party AI providers like Google and Anthropic, expanding beyond its current integration with OpenAI's ChatGPT.

Meta set to throw billions at startup that leads AI data market
Meta set to throw billions at startup that leads AI data market

Time of India

time2 hours ago

  • Time of India

Meta set to throw billions at startup that leads AI data market

Three months after the Chinese artificial intelligence developer DeepSeek upended the tech world with a model that rivaled America's best, a 28-year-old AI executive named Alexandr Wang came to Capitol Hill to tell policymakers what they needed to do to maintain US dominance. The US, Wang said at the April hearing, needs to establish a 'national AI data reserve,' supply enough power for data centers and avoid an onerous patchwork of state-level rules. Lawmakers welcomed his feedback. 'It's good to see you again here in Washington,' Republican Representative Neal Dunn of Florida said. 'You're becoming a regular up here.' Wang, the chief executive officer of Scale AI , may not be a household name in the same way OpenAI's Sam Altman has become. But he and his company have gained significant influence in tech and policy circles in recent years. Scale uses an army of contractors to label the data that tech firms such as Meta Platforms Inc. and OpenAI use to train and improve their AI models, and helps companies make custom AI applications. Increasingly, it's enlisting PhDs, nurses and other experts with advanced degrees to help develop more sophisticated models, according to a person familiar with the matter. Put simply: The three pillars of AI are chips, talent and data. And Scale is a dominant player in the last of those. Now, the startup's stature is set to grow even more. Meta is in talks to make a multibillion-dollar investment in Scale, Bloomberg News reported over the weekend. The financing may exceed $10 billion in value, making it one of the largest private company funding events of all time. The startup was valued at about $14 billion in 2024, as part of a funding round that included backing from Meta. In many ways, Scale's rise mirrors that of OpenAI. Both companies were founded roughly a decade ago and bet that the industry was then on the cusp of what Wang called an 'inflection point of AI.' Their CEOs, who are friends and briefly lived together, are both adept networkers and have served as faces of the AI sector before Congress. And OpenAI, too, has been on the receiving end of an 11-figure investment from a large tech firm. Live Events Scale's trajectory has shaped, and been shaped by, the AI boom that OpenAI unleashed. In its early years, Scale focused more on labeling images of cars, traffic lights and street signs to help train the models used to build self-driving cars. But it has since helped to annotate and curate the massive amounts of text data needed to build the so-called large language models that power chatbots like ChatGPT. These models learn by drawing patterns from the data and their respective labels. Discover the stories of your interest Blockchain 5 Stories Cyber-safety 7 Stories Fintech 9 Stories E-comm 9 Stories ML 8 Stories Edtech 6 Stories At times, that work has made Scale a lightning rod for criticisms about the unseen workforce in places such as Kenya and the Philippines that supports AI development. Scale has faced scrutiny for relying on thousands of contractors overseas who were paid relatively little to weed through reams of online data, with some saying they have suffered psychological trauma from the content they're asked to review. In a 2019 interview with Bloomberg, Wang said the company's contract workers earn 'good' pay — 'in the 60th to 70th percentile of wages in their geography.' Scale AI spokesperson Joe Osborne noted that the U.S. Department of Labor recently dropped an investigation into the company's compliance with fair labor regulations. Scale's business has evolved. More tech firms have begun to experiment with using synthetic, AI-generated data to train AI systems, potentially reducing the need for some of the services Scale historically provided. However, the leading AI labs are also struggling to get enough high-quality training data to build more advanced AI systems that are capable of fielding complex tasks as well as, or better than, humans. To meet that need, Scale has increasingly turned to better-paid contractors with graduate degrees to improve AI systems. These experts participate in a process known as reinforcement learning, which rewards a system for correct answers and punishes it for incorrect responses. The experts who work with Scale are tasked with constructing tricky problems – tests, essentially – for the models to solve, according to a person familiar with the matter who asked not to be named because the information is private. As of early 2025, 12% of the company's pool of contributors who work on the process of improving these models had a PhD in fields such as molecular biology and more than 40% had a master's degree, law degree or MBA in their field, the person said. Much of this process is aimed at companies that want to use AI for medical and legal applications, the person said. One area of focus, for example, is getting AI models to better answer questions regarding tax law, which can differ greatly from country to country and even state to state. Bets like those are driving significant growth for the company. Scale generated about $870 million in revenue in 2024 and expects $2 billion in revenue this year, Bloomberg News reported in April. Scale has seen demand for its network of experts increase in the wake of DeepSeek, the person familiar with the matter said, as more companies invest in models that mimic human reasoning and carry out more complicated tasks. Scale has also deepened its relationship with the US government through defense deals. Wang, a China hawk, has cozied up to lawmakers on the hill who are concerned about China's ascendance in AI. And Michael Kratsios, a former executive at Scale, is now one of President Donald Trump's top tech aides, helping to steer US policy on AI. For Meta, partnering more deeply with Scale may simultaneously help it keep pace with AI rivals like Google and OpenAI, and also help it build deeper ties with the US government at a time when it's pushing more into defense tech. For Scale, a tie-up with Meta offers a powerful and deep-pocketed ally. It would also be a fitting full-circle moment for Wang. Shortly after launching Scale, Wang said he was asked by one venture capitalist when he knew he wanted to build a startup. In response, Wang said he 'rattled off some silly answer about being inspired by The Social Network,' the film about the founding of Facebook.

Apple researchers say models like ChatGPT o3 look smart but collapse when faced with real complexity
Apple researchers say models like ChatGPT o3 look smart but collapse when faced with real complexity

India Today

time2 hours ago

  • India Today

Apple researchers say models like ChatGPT o3 look smart but collapse when faced with real complexity

They may talk the talk, but can they truly think it through? A new study by Apple researchers suggests that even the most advanced AI models like ChatGPT o3, Claude, and DeepSeek start to unravel when the going gets tough. These so-called 'reasoning' models may impress with confident answers and detailed explanations, but when faced with genuinely complex problems, they stumble – and sometimes fall flat. advertisementApple researchers have found that the most advanced large language models today may not be reasoning in the way many believe. In a recently released paper titled The Illusion of Thinking, researchers at Apple show that while these models appear intelligent on the surface, their performance dramatically collapses when they are faced with truly complex study looked at a class of models now referred to as Large Reasoning Models (LRMs), which are designed to "think" through complex tasks using a series of internal steps, often called a 'chain of thought.' This includes models like OpenAI's o3, DeepSeek-R1, and Claude 3.7 Sonnet Thinking. Apple's researchers tested how these models handle problems of increasing difficulty – not just whether they arrive at the correct answer, but how they reason their way The findings were striking. As problem complexity rose, the models' performance did not apparently degrade gracefully – it collapsed completely. 'They think more up to a point,' tweeted tech critique Josh Wolfe, referring to the findings. 'Then they give up early, even when they have plenty of compute left.' Apple's team built custom puzzle environments such as the Tower of Hanoi, River Crossing, and Blocks World to carefully control complexity levels. These setups allowed them to observe not only whether the models found the right answer, but how they tried to get found that:-At low complexity, traditional LLMs (without reasoning chains) performed better and were more efficient-At medium complexity, reasoning models briefly took the lead-At high complexity, both types failed completelyEven when given a step-by-step algorithm for solving a problem, so that they only needed to follow instructions, models still made critical mistakes. This suggests that they struggle not only with creativity or problem-solving, but with basic logical execution. The models also showed odd behaviour when it came to how much effort they put in. Initially, they 'thought' more as the problems got harder, using more tokens for reasoning steps. But once a certain threshold was reached, they abruptly started thinking less. This happened even when they hadn't hit any computational limits, highlighting what Apple calls a 'fundamental inference time scaling limitation.'advertisementCognitive scientist Gary Marcus said the paper supports what he's been arguing for decades: these systems don't generalise well beyond their training data. 'Neural networks can generalise within a training distribution of data they are exposed to, but their generalisation tends to break down outside that distribution,' Marcus wrote on Substack. He also noted that the models' 'reasoning traces' – the steps they take to reach an answer – can look convincing, but often don't reflect what the models actually did to reach a State University's Subbarao (Rao) Kambhampati, whose previous work has critiqued so-called reasoning models, was also echoed in Apple's findings, points out Marcus. Rao has shown that models often appear to think logically but actually produce answers that don't match their thought process. Apple's experiments back this up by showing models generate long reasoning paths that still lead to the wrong answer, particularly as problems get the most damning evidence came when Apple tested whether models could follow exact instructions. In one test, they were handed the algorithm to solve the Tower of Hanoi puzzle and asked to just execute it. The models still failed once the puzzle complexity passed a certain conclusion is blunt: today's top models are 'super expensive pattern matchers' that can mimic reasoning only within familiar settings. The moment they're faced with novel problems – ones just outside their training data – they findings have serious implications for claims that AI is becoming capable of human-like reasoning. As the paper puts it, the current approach may be hitting a wall, and overcoming it could require an entirely different way of thinking about how we build intelligent systems. In short, we are still leaps away from AGI.

DOWNLOAD THE APP

Get Started Now: Download the App

Ready to dive into the world of global news and events? Download our app today from your preferred app store and start exploring.
app-storeplay-store