• Failory
  • Posts
  • The Cost of Remembering

The Cost of Remembering

How Mem’s $99 AI assistant couldn’t sustain the business.

Hey, it’s Nico.


Welcome to another Failory edition. This issue takes 5 minutes to read.
Here are the 3 most important things of today’s issue:

Today’s newsletter is sponsored by Mixmax, the the Gmail tool that helps founders track email opens and automate follow-ups so investor and prospect conversations don’t go cold.

Never Let an Investor or Prospect Go Quiet Again: Mixmax for $1/Month AD

If you’re a founder running your own sales or fundraising, you know how fast conversations go cold. An investor asks for your deck, you send it, and then nothing. A prospect seems interested, then goes dark after your first email.

Mixmax fixes that from inside Gmail. It works where you already spend your day, so there’s no new tool to learn. You can see exactly who’s opened your emails and set up automated follow-up sequences that nudge people when they don’t reply. The sequence stops on its own the moment they respond. You no longer have to track threads in a spreadsheet or remember who to chase.

Right now, startups can get Mixmax Pro for just $1 per seat/month for the first six months, including 1,500 sequence recipients per month. Use code STARTUPS to claim the offer. It’s billed monthly and moves to $105 per seat/month after six months.

This Week In Startups

🔗 Resources

Want to show up when buyers ask ChatGPT what to buy? HubSpot for Startups’ free guide walks you through your first ChatGPT Ads campaign.*

No technical co-founder? Check SquadPrime. 2 selected founders are getting their MVP built free, for an honest review.*

📰 News

Anthropic blows past $2 trillion in Nasdaq debut

OpenAI releases GPT-6.2 Astra and Sol

Google gives Gemini the ability to use your Mac for you

💸 Fundraising

Isomorphic Labs, a startup that designs and develops medicines with AI, raised $5B in Series C funding

Genesis AI, a startup that builds general-purpose robots for industrial work, raised $500M in Series A funding

Model ML, a startup that builds AI agents for banking and consulting teams, raised $250M in Series B funding

BrightAI, a startup that uses AI sensors and robots to monitor utility infrastructure and industrial equipment, raised $97M in funding

*sponsored

Fail(St)ory

The $99 Second Brain

Mem built an AI notes app that helped people capture information and find it when they needed it. On November 16, it announced its wind-down, with the service closing January 15, 2027.

Its move into proactive assistance exposed a difficult tension: keeping information easy to save while charging enough for an AI that kept working in the background.

What Was Mem:

A meeting ends, another starts, and the useful thing someone said disappears into a document you might never open again. Mem wanted to make that information available without requiring a filing system.

Users could write notes, save web content or record meetings. AI helped connect related material and retrieve it through search or questions about what they’d saved.

For consultants, founders and executives, the appeal was having past conversations available during the next piece of work. The archive could become more useful as it grew.

Mem sold monthly software subscriptions. Its free plan limited usage, while Mem Pro cost $12 a month and gave users more room to save notes and work with their information.

It later introduced Mem Agent, which used saved notes and calendar context to track unfinished work and send reminders. A commitment mentioned in a meeting could become something Mem followed up on afterward.

The Agent initially came with Mem Proactive, a $99-a-month plan. Co-founder and CEO Kevin Moody explained that the assistant kept reasoning and reprioritizing in the background, adding computing costs even when the customer wasn’t actively using it.

In August, Mem replaced that lineup with usage-based plans. Proactive features became available across the tiers, including free, with monthly allowances for notes, messages and other activities. Paid plans started at $9, the new Pro plan cost $29, and the highest tier cost $199. Customers on older plans kept them at the time of the change.

When announcing the closure, Moody said repeated product and pricing changes had never produced a sustainable business at the intended scale.

The Numbers:

  • 🚀 Started: Late 2019

  • 💰 Disclosed rounds: Approximately $29.1M combined

  • 💳 Proactive plan: $99/month at launch

  • 🔚 Scheduled closure: January 15, 2027

Reasons for Failure: 

  • AI costs made flat pricing difficult: Moody said flat plans subsidized subscribers using the product 10–100 times as much as others, while the proactive assistant kept consuming compute in the background. Covering that work pushed Mem toward higher prices or tighter allowances. That likely made it harder to offer an affordable subscription while earning enough from its heaviest users.

  • Similar features became available in existing tools: Notion added AI meeting notes and search across connected apps. Mem itself let Claude search, summarize and update its notes. Users could get more of the job done through tools they already used.

  • Winning users required changing established habits: Mem's value depended on having the user's information. Its own onboarding guide asked people to build a capture habit or import existing notes. Customers had to move part of an archive or start saving their work in Mem, then decide whether better retrieval justified switching. That slowed the conversion of curious users into regular paying customers.

Why It Matters: 

  • Set AI pricing around heavy users’ costs, not average usage.

  • Prove why customers should pay when their existing tools add similar features.

  • Show value before asking users to migrate their data or change habits.

Trend

No AI Slop

You’ve probably had this land in your inbox: a polished AI-generated report that looks like a week’s work. Then you read it and realize you still have to check the numbers, untangle the assumptions, and decide what any of it means for your business.

Someone can look very productive while passing you the problem they were supposed to solve. With a few prompts, they can give you a document that takes the rest of your afternoon to evaluate.

That exchange has been part of working with AI for a while. Last week, speakers and attendees at AI Engineer Code Summit put together a No AI Slop manifesto. It caught my attention because it’s an attempt to agree on some basic AI etiquette, now that flooding people with stuff takes almost no effort.

Why it Matters

  • Individual time savings can disappear into the work colleagues spend checking, interpreting, and finishing an AI-generated delivery.

  • A finished-looking document can earn its sender credit before anyone establishes whether it helps solve the task.

  • Shared standards give teams a way to ask for better work and make the sender’s responsibilities explicit.

Too Much to Read

Developer Simon Willison recently ran into a familiar problem with his coding agents. They were writing accurate, comprehensive descriptions of proposed code changes. There was simply too much to read, even when the agent producing it was his own.

That’s an awkward side effect of AI productivity: we can produce things faster than we can evaluate them. Every extra page, proposal, or code change asks for someone’s attention. Making it takes less effort; deciding whether it’s useful can still take an afternoon.

A study updated this month shows how that imbalance can grow. Across nearly 200,000 proposed code changes at one software company, the volume of changes grew 3.1 times while the number of reviewers grew 1.5 times. Each reviewer ended up handling roughly twice as many changes.

The company shipped more code, and researchers found no clear decline in the quality measures they tracked. Those gains came with a growing pile of work to check. That’s the catch with celebrating how much AI helps us produce: some of the effort we save can become work for whoever receives the result. A five-minute prompt can leave someone else with an hour of figuring out whether any of it makes sense.

An Agile Manifesto for The AI Era

In 2001, the Agile Manifesto set out shared values for software development, emphasizing working software, collaboration, and adapting to change.

The No AI Slop authors are borrowing that approach for AI-assisted work. Their principles ask people to understand what they produce, take responsibility for its quality, and respect the time of whoever receives it.

Some projects have already written their own rules. The Open Source Robotics Foundation’s policy, updated this month, allows AI-assisted code but requires contributors to understand and verify what they submit. They must write the pull-request description themselves.

The policy explains that barely verified contributions push debugging and validation onto maintainers. A contributor who cannot explain their proposed change is effectively asking someone else to finish understanding it.

The Trend

AI makes it easy to look like you’ve solved a problem while leaving the hard part to somebody else. You can generate a detailed answer, send it along, and consider your part done. Whoever receives it still has to figure out whether it’s correct, useful, or worth reading. The output looks finished long before the thinking is.

I think that’s one of the diseases of the AI era: we can each feel more productive while making increasing demands on everyone else’s attention. The manifesto is an attempt to put some boundaries around a habit we already recognize. Its appeal says something about how exhausting that habit has become. We’ve all been handed something that took seconds to generate and far longer to make sense of.

Help Me Improve Failory

How useful did you find today’s newsletter?

Your feedback helps me make future issues more relevant and valuable.

Login or Subscribe to participate in polls.

That's all for today’s edition.

Cheers,

Nico