Happy Wednesday, Lattice Community!
We’ll dissect big stories in AI this week, provide insights to keep you informed in your industry, and share updates on the work we’re doing at Lattice Partners. We’re an AI implementation company that helps businesses with AI strategy and custom solutions.
This week’s TL;DR:
• Personal agents are having a moment
• People don’t realize how cheap AI is becoming and what this means for the future
• Agents are starting to transact on our behalf, but are they safe?
Personal agents have arrived. Should you be using one? And which one?
Meta launched Muse on September 8th, a personal AI agent, becoming the top free app in the App Store with over 2.5 million downloads in the first two weeks.
Instinct’s personal agent isn’t even technically released to the public yet, but it has a $10 billion valuation already.
And just yesterday, OpenAI announced the release of Dots at its DevDay conference.
Takeaways
1. How to know which personal agent is right for you:
Instinct operates like a text convo with a friend. It’s one thread where you go back and forth - it can handle bookings, email, and browse on your behalf.
Meta’s Muse also runs in a chat interface, but within its own app instead, connected to the Facebook and Instagram ecosystem. And if you miss your Tamagotchi days, don’t worry. Meta announced Muse Charm at its Connect conference, a cute keychain-sized device for your Muse agent’s avatar that allows you to chat with it, hands-free.
With Dots, you start with a “primary Dot,” name it, and it works in the background on tasks that you never have to direct. OpenAI says ideally you will have a whole team of Dots operating on your behalf, with “specialist Dots” for subjects like accounting or legal research. And unlike Muse, Dots is not free - it’s locked to Pro, Business Premium and Enterprise plans only.
2. Now you might never have to sit on hold with customer service again.
Let’s say you have your personal agent book you a reservation for a Friday night. It creates a plan, figures out which platforms, accounts, and tools it’ll need to execute. It opens its own browser and begins taking action - filling in forms, messaging, paying, and now even calling on your behalf!
3. These don’t replace Claude or ChatGPT for work.
While these agents are great at casual consumer tasks, they’re not meant for work use yet. They lack the tools, connectors, and raw model quality that other work tools have.
AI became 725x cheaper in just 18 months.
In January 2025, OpenAI’s o3 model received a 75% score on a PhD-level science exam, costing 30 cents per question. In only 18 months, GPT-5.6 Luna got the same score for four hundredths of a penny.
Takeaways
1. Imagine if a $50,000 car today dropped to $69 just 18 months later. Everyone would buy one, right?
It’s giving…Jevons paradox. You’d think that if something gets 725 times cheaper, total spending would drop because the cost is almost nothing now. But in reality, when something like AI is dramatically cheaper, usage goes up because everyone is finding thousands of new use cases for it. The same thing happened with electricity, gas and internet bandwidth.
2. This is why the personal agent boom is happening.
If AI was as expensive as 9 months ago, we wouldn’t have personal agents texting us all day, checking our calendars or browsing the internet. As model prices decrease, companies like Meta can offer something like Muse for free, broadening access.
3. Companies have to price this in.
Harvey, a legal AI company, charges law firms a flat fee per seat and was paying OpenAI and Anthropic for their models. As models have improved, lawyers are now using it up to 20x more, and Harvey’s costs have climbed dramatically. Instead of paying AI labs way more money for increased usage, it actually became more cost-efficient for Harvey to fine-tune their own cheaper, open-source model in-house.
We dive into all these topics more on the Lattice Partners podcast, the AI Playbook, linked here.
Agents don’t have their own payment systems…but they might need one.
Takeaway
We’re lacking agent-payment infrastructure.
Agents are technically making purchases on behalf of humans right now, but there’s nothing built yet to make it safe for an agent to actually talk to your bank or your credit card company.
BlackRock argues that stablecoins might be the one part of crypto stable enough to work as “machine-native money,” a currency built for machines to transact with each other, including AI agents.
For agents to transact at large volumes, there needs to be infrastructure built to ensure these payments are legit. Right now, they still require user-initiated action to verify credit card numbers or pay through “secure links.” Traditional payment rails were not built for agents - they were built for humans.
Maybe don’t give your credit card information to your personal agent…yet.
Co-founder Christian Yeghnazar’s experience with Instinct
Christian used Instinct to try to purchase concert tickets. He was offered a choice between typing his credit card information into the chat or using a “secure link.” He backed out because it didn’t feel safe enough. Personal agents are quite new, so Christian recommends that you first try out:
• something low-stakes and reversible, not a big purchase
• have your agent research, compare, or find you a great deal on different options - don’t pull the trigger too fast!
• make sure you stop before letting the agent complete a real purchase or entering payment info until you’ve read the fine print - what permissions are you granting it?
This week at Lattice Partners
Our team visited our client onsite in Tampa, Florida!
Martin and Colin visited our client’s facilities to gain context around their business, identify gaps in their processes, and propose where Lattice Partners could be helpful. We left with lots of ideas for standardizing how new parts get added to their systems and tracked from there. We also trained their team on Claude and the custom tool we built for them.
See you next Wednesday!
Cheers,
Lattice Partners
