The rise of agents, the affordability shift, and next gen AI videos [Horizon 01 - Issue #1]

As always, this was a busy week for AI developments. However, certain things are crystal clear.

AI is no longer just a tool—it’s becoming an active operator in business workflows.

The rise of autonomous agents like OpenAI’s Operator and Convergence’s Proxy means AI can now execute tasks, not just suggest them. For SMBs, this is a game-changer: routine operations like data entry, scheduling, and lead generation can be delegated, freeing up time for strategy and growth.

AI is becoming both more powerful and more affordable.

Microsoft’s “Think Deeper” Copilot upgrade and OpenAI’s cost-efficient o3-mini signal the affordability shift. This means SMBs can now leverage enterprise-grade AI capabilities without enterprise budgets. Whether automating customer inquiries, optimizing marketing strategies, or enhancing sales outreach, AI is leveling the playing field.

Content creation is multi-modal now, thanks to new AI tools.

DeepSeek’s Janus Pro and (eventually) ByteDance’s OmniHuman are reshaping content creation. AI-generated images, videos, and voice content are becoming indistinguishable from human-produced media, making high-quality marketing materials accessible to everyone. SMBs can now compete with larger brands in digital storytelling without expensive production teams.

AI isn’t just about automation—it’s transforming decision-making.

AI-powered analytics can now sift through massive datasets, extracting actionable insights that help businesses make smarter, data-driven decisions. Tools like Nowadays and Falcon AI show that even high-touch tasks, from event planning to content moderation, can be streamlined.

AI is no longer optional. Businesses integrating AI into their marketing operations are posed to gain efficiency, cost savings, and competitive advantage. Those who delay risk falling behind. We can help you identify the right AI strategies tailored to your business needs if you're unsure where to start. The AI-driven future is here—let’s make sure you’re ready for it.

OpenAI’s Operator: A Glimpse of Autonomous Agents

OpenAI made waves with the early preview of Operator, one of its first truly autonomous AI agents. Unlike a standard chatbot that only outputs text, Operator can take actions online on your behalf. It uses a “computer-using agent” model, meaning it can navigate websites through a browser, click links, fill forms, and execute tasks like a human would. In other words, you can give Operator a high-level task – “book my team’s flights to the LA conference” or “order office snacks for next week” – and it will handle the web browsing and form-filling necessary to get it done. If it hits a snag or needs sensitive info (like a payment), it hands control back to you to confirm. Currently, Operator is only available to U.S. users on ChatGPT’s $200/month Pro tier, but OpenAI plans to expand it to broader plans and eventually the free version.

Why it matters:

This is a peek into the future of “AI agents” doing repetitive digital chores. Marketers could eventually use Operator to pull weekly web analytics reports or update campaign dashboards across different sites automatically. Sales reps might have it scour several websites for lead information or fill out CRM forms. Customer support teams could offload routine web-based tasks (like checking order statuses across systems) to an agent. Keep an eye on Operator – while it’s in early days, its ability to “take control of a web browser and perform tasks for you” signals a coming era of hands-free automation for all those little tasks that eat up your time.

Convergence “Proxy” – An AI Assistant for Everything (Now with a Free Tier)

Not to be outdone, a startup called Convergence launched Proxy, an automated AI assistant designed to handle the drudgery of online work. Much like Operator, Proxy can

“interact with the web on your behalf, navigating sites, forms, databases… in much the same way as a human.”

It’s built to function as both a reliable personal assistant for life admin and a skilled colleague at work. What’s unique is Proxy’s focus on automation and repetition: you can “set and forget” recurring tasks and it will repeat them on schedule, getting more efficient each time. For example, you could instruct Proxy to send you a synthesized report of your competitor’s website updates every Friday, or have it routinely pull sales leads from a database each week without prompting. Crucially, Proxy is available globally right now – they even offer a free tier and affordable plans starting at $20/month, which is a “radical new benchmark” for this kind of AI service.

How you can use Proxy:

Think about any repetitive digital task you do: compiling data into a spreadsheet, checking multiple sites for updates, booking appointments, or sending routine emails. Those are great candidates to delegate to Proxy. For instance, marketers can automate weekly social media performance reports by letting Proxy gather metrics from each platform. Sales teams might use it to automatically scrape and pre-qualify leads from public sources. And support teams could offload routine ticket triage or FAQs – Proxy could copy relevant info from a CRM or knowledge base and even draft initial responses. Since it’s customizable and will ask for guidance if it’s unsure, you remain in control. This launch is a sign that 2025 is set to be a big year for AI assistants, with Proxy showing that powerful agents don’t have to break the bank.

Microsoft’s “Think Deeper” Upgrade in Copilot

If you’re using Microsoft’s Copilot, prepare for a power-up. Microsoft’s AI CEO Mustafa Suleyman announced that all Copilot users now have free access to Think Deeper, which integrates OpenAI’s advanced o1 reasoning model across tasks in math, science, coding, and more. This comes after feedback about earlier limitations, and it effectively gives every user of Copilot a dose of deeper analytical capability – at no extra cost. What can it do? Think Deeper is touted as

“truly magical”

for generating in-depth plans and advice. For example, you can brain-dump a complex project or campaign idea, and watch Copilot churn out a detailed step-by-step execution guide.

How to leverage it:

Marketing managers might use it to outline multi-channel campaign plans. Sales professionals can brainstorm account strategies or get coaching on complex deal scenarios. Support teams could have it generate structured troubleshooting guides or training plans. With richer reasoning now built into Copilot, you can ask more ambitious questions – and get more structured, insightful answers – to drive your projects forward.

OpenAI o3-mini: Smarter Reasoning at Half the Cost

On the AI model front, OpenAI has released o3-mini, its newest “reasoning series” model that’s both fast and cost-efficient. How cost-efficient? It’s priced at roughly $1.10 per million input tokens and $4.40 per million output tokens – less than half the price of the comparable GPT-4 model (aka “GPT-4o”). Despite the lower cost, o3-mini isn’t a lightweight in capability. OpenAI says this model

“advances the boundaries”

of smaller, efficient AI and is particularly strong in science, math, and coding tasks. It supports features like function calling and structured outputs, and even comes with adjustable “reasoning effort” levels (low, medium, high) so developers and users can balance speed vs. depth. Another big deal: o3-mini can search for up-to-date information and provide answers with linked sources, making it useful for research or any task that benefits from current data.

For the first time, OpenAI is also making an advanced reasoning model available to free ChatGPT users.

You can switch ChatGPT to the “Reason” mode to try o3-mini’s capabilities.

Why marketers and sales professionals should care:

Cheaper and smarter means you can do more with AI without blowing the budget. Teams using ChatGPT or building AI into their workflows can now afford to scale up those experiments – whether it’s generating long-form content (o3-mini can output up to 100k tokens in one go, far more than previous models), analyzing large datasets or chat logs, or powering an AI assistant for customers. For example, a sales ops team might use o3-mini via the API to auto-summarize every sales call recording overnight, giving reps a morning briefing. Marketing teams could feed in market research documents and ask for strategic insights or campaign ideas. Customer support could integrate o3-mini into chatbots to get more accurate and context-aware answers, thanks to its improved reasoning. All at a fraction of the previous cost. It’s a win-win: higher AI adoption without needing a higher budget.

DeepSeek’s Janus Pro Multimodal Leap

Chinese startup DeepSeek grabbed headlines this week by unveiling Janus Pro, a major upgrade to its multimodal AI system Janus. This new model version has been “completely overhauled” with refined training methods, a much larger dataset, and a bigger model size. Under the hood, Janus Pro ingested 90 million additional examples for understanding images (everything from diagrams to memes) and even used 72 million synthetic images(some generated by Midjourney). The result is a model available in up to 7 billion parameters, which shows notably better performance in both image understanding and generation tasks than its predecessor. In fact, in one benchmark (GenEval for image generation quality), Janus Pro scored 80% accuracy versus the old model’s 61%, even

outperforming OpenAI’s DALL-E 3

on that test. While it’s not beating the very top-tier models in every aspect (and it still has a constraint of 384×384 image outputs for now), Janus Pro demonstrates that cutting-edge AI isn’t confined to the usual players.

Why it’s relevant:

For one, DeepSeek’s rapid advances (its earlier R1 model last year was dubbed a potential “Sputnik moment” in AI) mean more competition and options in the AI landscape. Marketers and creators could soon have open-access multimodal tools that approach the quality of big corporate models – giving you more freedom to generate images or interpret visual data without vendor lock-in. In practical terms, a marketing team might experiment with Janus Pro (which is available on GitHub and Hugging Face) to generate product images or social media visuals in-house, keeping sensitive data off third-party platforms. Or a support team could use its image understanding to analyze screenshots from users (think: a customer sends a snapshot of an error, and an AI model like Janus Pro could help interpret it). At the very least, DeepSeek’s progress pushes everyone forward – even OpenAI slashed prices (see o3-mini above) to keep up.

The takeaway: multimodal AI is advancing fast, and even smaller players are delivering tools that businesses can potentially leverage for creative and analytic tasks.

OmniHuman Preview: Turning a Photo and Audio into Video

Ever wish you could create a polished marketing or training video without a camera crew? A new research breakthrough suggests that future might be closer than you think. OmniHuman-1, developed by a team at ByteDance (yes, TikTok’s parent company), is a model that can generate. Imagine snapping a photo of yourself or an actor, feeding it an audio recording of a speech, and getting back a video of that person delivering the message – with natural movements, expressions, even gestures that match the speech. That’s essentially what OmniHuman achieves in the lab. The researchers introduced a new training strategy to make this possible, and it “significantly outperforms existing methods”, producing

“extremely realistic human videos based on weak signal inputs, especially audio.”

It handles any aspect ratio (portrait, full-body, etc.) and maintains realism in motion, lighting, and texture details, according to the project page. Now, OmniHuman isn’t publicly available yet (the creators note they are not offering downloads or services at this time), so you can’t deploy it in your marketing stack tomorrow. But it’s important to know what’s on the horizon.

Potential use cases:

Marketers could eventually generate spokesperson videos or personalized ads without a lengthy video shoot – for instance, creating dozens of variant video ads with different voiceovers and having an AI “speak” them in a single actor’s likeness. Sales teams might craft custom video messages for key clients, with a virtual representative delivering a tailored pitch. Customer success and training departments could produce how-to videos or tutorials with just a script and a photo of a coach or agent. There are of course ethical and brand considerations (deepfake technology needs to be used responsibly), but the capability is rapidly growing. Start thinking about how your team might leverage quick, AI-generated video content if and when a tool like OmniHuman becomes available – it could radically reduce the cost and time of video production while enabling a new level of personalization.

AI Plans Your Next Event: “Nowadays”

Planning corporate events – from team offsites to client workshops – usually means hours of emails and calls to venues, right? Nowadays, a newly launched AI copilot for event planning, is here to change that. This product made a splash on Product Hunt (ranked the #1 product of the day on Jan 28, 2025) for good reason. It “takes the hassle out of organizing corporate events” by handling the legwork of contacting venues and even negotiating prices or terms. All you do is input the basic event details (dates, city, size, preferences) and let Nowadays do the outreach. It will email or call venues, gather proposals, and come back with options for you. Essentially, it’s like having a virtual event planner assistant available 24/7.

Why it’s useful:

Marketers often juggle event planning for product launches or user conferences; sales teams organize networking dinners or partner events; and customer success might plan training seminars. In all cases, coordinating with venues is tedious. By offloading that to an AI, your team can focus on the content and goals of the event rather than the logistics. Early users report significant time saved and quicker turnaround in locking down venues.

If you have an event on the horizon, give Nowadays a try for the venue-sourcing phase. It’s an example of AI tackling a very human-intensive task (negotiating with vendors) and shows how narrowly focused AI solutions can free you up for higher-level work. Plus, since it was a top-rated launch, the tool comes from a credible source and has good early feedback from the community.

Keeping Communities Safe: New NSFW Image Detection

For community managers and support teams, moderating user-generated content is a never-ending concern – especially images that might be inappropriate or harmful. Enter Falcon AI’s latest NSFW image detection model, a fine-tuned Vision Transformer that is specifically trained to classify images as “normal” or “NSFW” with a high degree of accuracy. This model was built on a curated dataset of ~80,000 images across those two categories, learning the nuanced differences to reliably catch explicit content. The result, as the model creators put it, is

“a model that stands ready to contribute significantly to content safety and moderation, all while maintaining the highest standards of accuracy and reliability.”

In practice, developers or IT teams can integrate this open-source model into their platforms – it’s available on Hugging Face – to automatically scan user-uploaded images (in forums, social networks, comments, etc.) and flag or block those deemed NSFW before they ever reach the eyes of your broader community or customers.

How this helps you:

If your marketing involves any user-generated content campaigns (think photo contests or testimonials) or your company runs an online community, this AI tool can be a guardian. Instead of relying purely on manual moderation or older, less accurate filters, you can deploy a state-of-the-art detector to catch the majority of problematic images instantly. This not only protects your brand’s reputation by preventing a bad image from staying up too long, but also lightens the load on your community managers, who can focus on engaging with users rather than sifting out bad apples. Customer support teams on social media duty likewise can benefit – let the AI filter out the obvious NSFW images so they can concentrate on genuine customer issues. It’s a great example of AI aiding human teams to maintain a safe, welcoming environment online.

Insight: Marketing AI Adoption is No Longer Optional

All these tools underscore a broader trend: AI is swiftly becoming central to business functions. In fact, according to the 2024 State of Marketing AI Report (by the Marketing AI Institute and Salesloft), 66% of marketers say AI is either very important or critically important to their marketing success in the next 12 months. Over half of marketing teams are already piloting or scaling AI solutions in some form. What about the rest? The report hints that the remaining folks risk falling behind – or worse. As the AI Vibes newsletter quipped, that 34% of marketers not yet convinced might be the ones “getting laid off in the next 12 months” (a tongue-in-cheek way to stress how crucial AI proficiency is becoming). The key takeaway for all professionals – not just marketers – is clear: it’s time to start weaving AI into your workflows.

Actionable next steps:

If you’re new to AI, pick a small project where an AI tool might help (perhaps try one of the tools above). For a marketer, it could be experimenting with an AI copywriter or image generator for a low-stakes campaign. Sales teams might test an AI email assistant to draft follow-ups, and support teams could pilot a chatbot for common inquiries. Measure the results, learn, and iterate. The report shows that understanding of AI is mostly “intermediate” among marketers, meaning there’s plenty of room to learn – but also that you don’t need to be a PhD to start using these tools. In short, make 2025 the year you upscale your team’s AI capabilities. The tools are more accessible and affordable than ever (many are free or low-cost), and as this newsletter hopefully demonstrated, credible solutions abound for nearly every aspect of your go-to-market operations.