Ken Lu

At the end of this video, Claude is controlling Pontiac. This is the (very) first iteration of the 'Pontiac DSP chatbot', so it is running in Claude code which is a developers environment. It is a bit intimidating for the DSP buyer who is use to a glossy interface. Calm down! It will get "glossier". This is just the very early stages. Our version for Claude desktop comes out next week. That being said, you can see me controlling the platform with an LLM, and this is just the very, very beginning. Part of me realized that this is not going to replace the interface. It is just going to be a tool, or an assistant, who can accomplish some tasks very quickly, while also handling repetitive tasks with ease. Honestly, its hard to say where this is going. Its goin' somewhere though! Pretty incredible stuff. Pontiac Intelligence Programmatic Mechanics

54

7 Comments

3,144 new sites, $6,721 in MRR, and one hiring process that really tested my patience. Here’s the full breakdown of our public monthly review: 📈 3,144 new websites connected to WP Umbrella 🙌 99 new premium members joined 💸 MRR increased by $6,721 ⬇️ Churn: 2.6% 🛡️ Only 52 sites activated Site-Protect (powered by Patchstack) We also shipped 4 product updates: - Bulk updates are now manageable by tags - Updates and syncs are 5x faster - New Logs feature to track everything - Site-Protect reporting integrated in PDF reports Key Highlights 1) Update automation is almost ready. We’ve reworked everything around updates: faster, more reliable, and better organized. Since bulk management is why people log into WP Umbrella every day, this is great news. It’s something we should’ve done much sooner. 2) Hiring in the age of AI sucks. We received 600 applications in 5 days. Sent the technical test to about 40 people. Only 3 passed. I was shocked by how many low-effort, AI-generated applications we received from senior devs. I just don’t get it. What are people expecting? And I’m not even talking about junior profiles. The younger generation entering the job market is going to have a hard time if they don’t learn to think critically and stand out with real work. If everything you do looks like it was spit out by a tool, then I guess we can just replace you by a tool. 3) The price of being independent Last month, we turned down two major integration offers from big WordPress players. They could’ve brought us a ton of exposure and resources, but they also would’ve slowed us down on what matters most: improving the product for our existing users. We can afford to make calls like this because we’ve never raised money. And honestly, I’m really grateful for that. No fundraising. No flashy booths. Just a solid product and a team that genuinely cares. We’re not taking a summer break and we have big things planned for September. Thanks for sticking with us 💙

73

7 Comments

Revenue is the result. Process is the variable⚙️ You can’t optimize based on the outcome alone; you need metrics describing the process itself. The Problem: Measuring only eCPM (The Finish Line)🏁 The Solution: Measuring auction mechanics - bid responses and latency⏱️ To stop revenue leakage, you must monitor timeouts for both Prebid Client and Prebid Server. If you aren't tracking bid drop-offs, you're just guessing. Find your "revenue sweetspot" 🎯 Read more: https://lnkd.in/diVzgA7T #AdOps #Programmatic #HeaderBidding #Setupad

11

Initially tested on over 50,000 sites via a feature plugin, speculative loading has shown measurable performance improvements, including a ~1.9% boost in Largest Contentful Paint (LCP) passing rates. Read more 👉 https://lttr.ai/Akral #SpeculativeLoading #CloudflareSpeedBrain #LatestAdvancements #WebPerformance #UnlockBlazingFastExperiences #LargestContentfulPaint

1

How to put React components in your AI generated answers Vetted will soon have new product reviews on our product details pages with rich source snippets rendered directly in the answer text. To do this I needed to create a new data pipeline to allow parsing and rending JSX in markdown. I've open sourced the libraries. Here's how it works: https://lnkd.in/gXmyHa5X

15

More proof assets do not equal credibility. When proof is thin, don't make loud claims. Use a repeatable way to prove what you’re saying. The failure mode is simple. Startups makes claims without evidence and buyers can see the guess work As a PMM fix it with stacked proof One layer per core claim and standardizes it- claim → proof → measurement. Level 1: Mechanism proof Share screen flow, workflow diagram, demo steps that customers can relate to Level 2: Workflow proof Show steps removed, steps added, handoffs changed in the customer's process Level 3: Early signal proof Activation, time-to-first-value, usage frequency, retention trend, pilot completion. Level 4: Outcome proof Show potential win rate, conversion, cycle time, costs saved, revenue impact. Only if you can attribute it. Operating rule: Do not publish a claim above your highest proof level. If you don’t have outcomes, you still have options. Run a pilot to validate what is sticking and that can be the proof point Define the pilot user and workflow. Define one success metric tied to the workflow. Define one adoption metric tied to usage. Define the time window. Show what “good” looks like. At Awvent when we had no customer we applied this process with countless potential customer..refined it with every feedback...eventually landing our first customer ( always a sweet feeling) As early stages startup this is one way you stay credible while the product is still earning outcomes.

7

1 Comment

MULTI-COMPANY QSBS STACKING STRATEGY This is how founders quietly shield $50M–$100M+. Core Rule QSBS exclusion is per taxpayer, per company. Example Stack Company Exclusion Company A $10M Company B $10M Company C $10M Trust 1 $10M Trust 2 $10M Same founder. Same brain. Different taxpayers. ⸻ Advanced Stack (Founder Grade) 1. Founder owns shares 2. Founder gifts shares early to: • Spouse • Non-grantor trusts 3. Each entity gets its own QSBS clock 4. Exits occur after 5+ years 5. Exclusions stack cleanly ⚠️ Must be done before appreciation

1

Elementor vs Beaver Builder: The Features That Actually Matter for Agencies After my recent posts on page builder performance generated significant discussion, the most frequent pushback was: "Performance is one thing, but what about features?" Fair question. So I conducted a systematic feature analysis focused specifically on what agencies need for client work. The comparison covers: Themes and flexibility Performance and code output Long-term maintenance considerations Workflow integration Client handoff and training Full transparency: my testing confirmed my decision to standardize on Beaver Builder for agency work. But I wanted to document the complete feature comparison so agencies can evaluate the trade-offs for themselves. This completes my three-part research series documenting the testing process and results that informed my decision. Full feature analysis: https://lnkd.in/eQ-rAV5a For agency owners and development leads: What features are non-negotiable in your page builder choice? #WordPress #WebDevelopment #AgencyOperations

3

💡 A business is not a product. And a product is not a business. In this episode of The CTO Show With Mehmet Gonullu, Mohamed (Mo) Ahmed, Ph.D. broke it down simply: “A business is a machine—you feed it $2, and it gives you back $3. A product? That’s just a piece of the machine.” Too many engineers (my past self included) believe that building something elegant is enough. It’s not. Customers don’t pay for elegance—they pay for value. And value only matters when it turns into revenue. 🎧 In Episode 488, we dive into: — The product trap most technical founders fall into — Why mental conditioning matters more than pitch decks — And how Mohamed built, pivoted, and eventually exited his startup Listen on all podcast platforms + YouTube (links in comments)

9

1 Comment

US founders hiring Flutter in US time zones: your first paragraph decides who applies. If it's requirements and years, US seniors read it and think: "no chance." Lead with ownership instead: What ships in the first 90 days. What they own in production from day one. What they’ll have the authority to say no to. That’s the difference between a team add and a first mobile hire.

15

2 Comments

A Google Calendar invite can now execute code on your machine if you connect agents like Claude Code to your calendar. The vulnerability is real, but I think the framing can be better. This isn't a bug in the agents like Claude. It's a design problem we keep running into: giving agents access to read untrusted data AND execute privileged actions in the same session. I've been thinking about this a lot because I've been agentifying my life and business, and recently built an email triager that faces the same risk. It reads inbound emails from anyone (inherently untrusted, can contain prompt injection) and it reads my sensitive personal context to draft intelligent replies. That's a prompt injection playground. Here's my security model: the agent is locked down so it cannot send emails. It can label them and write drafts. That's it. I review drafts before sending. The agent has zero ability to create side effects I haven't approved. This also means I'm locking down WebFetch, and other tools that might inadvertently leak information. With the appropriate sandboxing + controls on data ingress and egress, I'm comfortable letting the agent run in the background while I sleep. This maps to recent DXT vulnerabilities reported. The attack works because a calendar reader (low-risk) can chain to a shell executor (high-risk) through MCP tool composition. If those were scoped so the calendar agent could only read and annotate, but never execute, the attack chain breaks. Prompt injection is like phishing: you will never stop all of it. But we don't secure humans by making them perfectly phishing-proof. We gate what actions they can take and monitor the risky ones. Same logic applies to agents. Secure the side effects, not the inputs.

14