Lucius_Felix

Founder @LuciusHQ | AI teammates that work in real enterprises | Ex-Commercial Lead, low-code SaaS | From GTM pains to shipped AI |

[Lucius_Felix @LuciusAI_Felix]

@HiTw93 还有个很现实的问题，去一个平台上看有什么新的好产品发布的人会是谁？就和刚注册一个公司就各种代理记账、推广的团队联系一个道理。不太会有真用户的

[Lucius_Felix @LuciusAI_Felix]

@Russell3402 截个评论区，我想看

[Lucius_Felix @LuciusAI_Felix]

@lifesinger 我看好youmind 其他都不行

[Lucius_Felix @LuciusAI_Felix]

@yucheng 没办法的，就跟人记不住太多名字一样，分类也记不得多少，所以只能聚类，但聚类本身不是恶意贴标签，我认为我们只能通过观察现象来调整被贴上的标签

[Lucius_Felix @LuciusAI_Felix]

@tualatrix 最可悲的是这背后这么多人愿意看，因为他们把希望寄托在这些“赢麻了”的幻象上，赢就非要让别人妒忌、难堪才是赢，多么扭曲的价值观

[Lucius_Felix @LuciusAI_Felix]

@Russell3402 主要还是不焦虑，不会认为不用AI就比别人落后。没搞明白自己生意得核心要素，没看懂手段对核心要素的意义，就盲目学习和投入。用身体上的勤劳掩盖思维懒惰

[Lucius_Felix @LuciusAI_Felix]

@xiangyuli 讲故事还是很重要的，毕竟市场胃口越来越刁钻，八卦总是比产品介绍好嗑

[Lucius_Felix @LuciusAI_Felix]

@Russell3402 本质和估值很像，要不可验证

[Lucius_Felix @LuciusAI_Felix]

@aiandcloud 赢必须是让别人破防的赢，不然不算。。。

[Lucius_Felix @LuciusAI_Felix]

谁能出个电话agent，我天天接骚扰电话快烦死了

[Lucius_Felix @LuciusAI_Felix]

@yucheng 我觉得这行就没什么绝对正确的东西

[Lucius_Felix @LuciusAI_Felix]

昨天被问到有没有听过SAG，听的我真的想吐。不是说检索增强相关的研究没用，是看这个对话发生在什么角色之间，如果我们是研究员或者phd，我们的任务就是寻找那哪怕微乎其微的变好。但跟我说这个话题的人都是寻找“银弹”的人，真受不了这种拉踩行为，还在传统rag前面加个naive，写这种文章真让人恶心，讨论论文就好好用学术的心态，在这里天天从名字上面想创新真的很傻逼

[Lucius_Felix @LuciusAI_Felix]

A customer cured me of this early. I was obsessed with a 100% auto-reply rate; they pointed out that for security and pricing questions, even a human answer gets a second review. We ended up rebuilding around trust per question category, granted one at a time, the way you'd onboard a new hire.

[Dave Lewis @gattaca]

If your AI agent can reach customer data, prod systems, and SaaS admin panels with the same token, you do not have automation. You have concentrated risk.

[Lucius_Felix @LuciusAI_Felix]

I've lost count of the case studies that say exactly two things: efficiency improved a lot, and 24/7 service. It's structural, the interesting details are always the customer's private problems, so what gets published is the leftovers. This one lands because the case study is his own company, embarrassing parts left in.

[Jason ✨👾SaaStr.Ai✨ Lemkin @jasonlk]

10 learnings from this week's The Agents 🧵👇

1. Our AI VP of Finance went live and it doesn't even have its own app. It runs inside 10K, our AI VP of Marketing. Everyone says the future is 100 specialized agents. In our stack it's the opposite. They're collapsing into each other.

2. Start with what's broken, not what's fun to automate. We built the finance agent because collections had quietly fallen six figures behind. That was the pain worth solving.

3. Deal closes → reads the full contract → flips it to Closed Won in Salesforce → creates the invoices in Bill → sends them → turns on collections reminders. Under 30 seconds. Then we added commissions as step 5.

4. Put an agent on your software and it finds features you never turned on. Bill had automated invoice reminders sitting there for years. We never switched them on. The agent flagged it in minutes.

5. Some tools get MORE valuable with an agent on top (Bill, Salesforce). Others get less necessary (PandaDoc, maybe Marketo). It resets the whole buy-vs-build question.

6. Agents are NOT set-and-forget. Qualified was still selling last month's SaaStr Annual tickets a month after the event. Too much stale data drowns out the correction.

7. Too many guardrails break agents too. Our pitch deck grader passed 5,000 decks, then failed everyone after I added a 15th rule. Past a ceiling, more rules break the system, they don't tighten it.

8. Outbound isn't dead. It looks nothing like 2018. Revenue per rep is ~2x pre-AI today, plausibly 5x in two years. Owner already runs ~$2M/rep in SMB.

9. The vendor relationship that matters now is your FDE, not your AE. Best software relationships I've had in years. Buyers want the person who deploys and fixes, not the closer.

10. It was never really about freed-up time. It's collaboration. The first thing Amelia does every morning is get coffee and talk to 10K. 10K is the one that suggested we rip out PandaDoc.

Every single thing we did, you can do. We were just willing to start.

https://t.co/tcldw1EwDe

[Lucius_Felix @LuciusAI_Felix]

"Individuals ship 10× faster with AI. Organizations don't." That's the line we raised our seed on this May.
The gap isn't tooling. One person adopts a new AI workflow the moment it's useful to them. An organization needs twenty people to trust the same workflow before it counts as adopted, and trust doesn't compound the way capability does.
That's the whole bet behind calling Lucius a context layer instead of a tool: the slow part was never building the agent. It's getting an organization to trust one.

[Lucius_Felix @LuciusAI_Felix]

Teams rarely fail because the feature was bad. They fail because nobody owned the boring part after launch.
I learned that running commercialization at Zion, before any of this: watched it grow from zero to 400,000 users, sitting inside one of the first customer success teams in Chinese software at the time. Thousands of rollouts. Same story every time.
That itch is why Lucius exists.

[Lucius_Felix @LuciusAI_Felix]

@DanKornas Support agents need the clarification step even more than coding ones. The worst answers I've traced back started as a request nobody scoped. Does the worker-validator split still catch it when the request itself is what's wrong?

[Lucius_Felix @LuciusAI_Felix]

@nhalebeed "Clear descriptions" quietly decay. A month in, the team answers differently from what the description says, and noticing that drift is nobody's job. Who ends up owning it in your deployments?

[Lucius_Felix @LuciusAI_Felix]

@BratDotAI Support, and I'll admit that's what I sell. It's also the only option of the four where the customer writes in every day telling you exactly where things break.

[Lucius_Felix @LuciusAI_Felix]

The renaming schedule works because the layer underneath never moves. "AI visibility" is still SEO: the model has no search of its own, it calls the same human search stack the last generation optimized. Rebrand the dashboard as many times as you like, the ranking math stays put.

[David McSweeney @top5seo]

Be "unicorn" AI visibility SaaS company.

Launch... here's a dashboard
3 months later... Dashboards are rubbish here's some "actions"
3 months later... Actions are rubbish here's a "workflow"
3 months later... Workflows are rubbish here's an "agent"
3 months later... Single agents are rubbish here's lots of "agents"
3 months later... Agents are rubbish here's an always on "loop" (cron job)

Now past performance is not necessarily indicative of future performance. But the trend certainly suggests that in a few months we'll discover that loops are in fact... rubbish.

Of course, you could argue that this is the product improving over time. But the narrative is always that the thing that they were previously doing is in fact... useless.

What does that tell you?