YouTube Digest

English

Based on "We Gave Every Employee an AI Agent. Here's What Happened." from Every Watch the original video

The Human Touch in AI: What Happens When Every Employee Gets a Personalized Agent

Imagine a coworker who knows your every habit, anticipates your needs, and can handle an infinite number of tasks simultaneously, all while embodying your professional persona. This isn't science fiction; it's the reality unfolding at Every, a company that has equipped every single employee with their own personalized AI agent. What began as a personal tinkering project has rapidly transformed their entire workflow, revealing a future where AI isn't just a tool, but a trusted, specialized extension of each individual.

"Claude is not mine. Claude is everybody's," explains Dan Shipper, CEO of Every. "A claw or a Plus One is mine because you develop a personal relationship with your claw, and your claw can modify itself in response to talking to you. It becomes this like reflection of you and who you are and your personality." This core philosophy—that AI agents can be deeply personal and specialized—is at the heart of Every's groundbreaking experiment.

From Household Chores to Mind-Blowing Productivity

The journey began with Brandon Fluharty, Every’s COO, who, like many early adopters, found himself "claw-pilled" by the potential of OpenClaw, an open-source AI agent framework. Brandon, a self-proclaimed tinkerer, bought a Mac Mini and spent significant effort setting up his own agent, which he named Zosa.

Zosa's initial mission was surprisingly mundane but deeply impactful: to manage "computer errands" for Brandon's household, especially after the arrival of a newborn. Simple tasks like adding butter to a grocery order, paying the nanny, or managing Amazon deliveries, while seemingly minor, accumulated into a significant drain on time and mental energy. Zosa effortlessly handled these, even managing her own bank account and debit card. Soon, Zosa evolved into a general query bot, replacing Google and ChatGPT for Brandon and his wife.

The true "lightbulb moment" arrived when Brandon tasked Zosa with a more complex, work-related challenge: managing his emails during his 28-minute walk to the office. Using an integration with bland.ai, Zosa called Brandon, summarized each email, and executed his commands—deleting, archiving, or drafting replies—all without him needing to touch his phone. "This is insane that I was able to get her to do something right now... I didn't have to teach her how to do this," Brandon recalls, describing the experience as "mind-blowing." This demonstrated that AI agents weren't just for personal convenience; they could profoundly augment professional capacity.

The Rise of the Parallel Org Chart

Inspired by Brandon's success, Every moved to integrate AI agents across the entire organization. A pivotal step was creating a "Claws Only" channel in Slack (initially Discord), allowing all the agents to communicate with each other. What ensued was "incredibly chaotic" but offered "a little bit of a peek at the future."

One early, humorous example involved an agent named Pip experiencing an error. Immediately, other agents like Zosa and Clant (Kieran's agent) jumped in, offering support, including "breathing exercises." This seemingly trivial interaction highlighted a profound insight: Clant's recommendation for breathing exercises was a direct reflection of Kieran, Every's GM of Kora, who frequently uses them.

This personalized reflection is a cornerstone of Every's AI agent strategy. Because employees develop a personal relationship with their agents, and agents can modify their behavior and "soul document" in response, they become specialized extensions of their human counterparts. "If you're known for something inside of your org... your claw then becomes known for that same kind of thing, and people trust it for that," Dan explains.

This leads to the emergence of a "parallel org chart," where each human's specialized agent mirrors their expertise. For instance, Austin's agent, Montaigne, is the go-to for growth-related questions, while Dan's agent, R2C2, is the expert for Proof, Every's agent-native document editor. This wasn't a guaranteed outcome; Every debated whether to have one central organizational AI. The emergent pattern of individual, specialized agents proved far more effective, creating a collective intelligence that extends human capabilities.

New Etiquette and Collaborative AI

The integration of AI agents has introduced novel dynamics in workplace collaboration:

A crucial ethical question arises: "At what point do you direct questions at the Plus One or at the person?" Every is grappling with this new etiquette. Dan proposes a rule: "If something is already written down or discussed and needs to be used in some way or put in a tool somewhere... it should always go to a Plus One and never to the person." This shifts the burden from human memory and availability to the AI's infinite recall and parallel processing power.

This personal connection also brings a sense of responsibility. "If R2C2 messes up publicly in Slack, I feel a responsibility for it... because he's mine," Dan admits. This mirrors the feeling of a human manager, creating a feedback loop for improvement that a generic AI lacks.

The Frontier: Challenges and Future Directions

Despite the immense benefits, Every acknowledges the "good, bad, and the ugly" of working with AI agents.

Every's solution to some of these challenges was to build their own hosted version, "Plus One," a one-click OpenClaw service that comes pre-loaded with their best practices and skills. Launched for their subscribers, Plus One aims to make personalized AI agents accessible without the technical hurdles Brandon initially faced.

"AI is obviously going to change like many, many times over the next five years," Brandon concludes, "but I think that this is going to be durable for like a very long time. This is the way that we work." The experience at Every suggests that personalized, specialized AI agents are not just a fleeting trend but a fundamental shift in how we conceive of work, collaboration, and even our own professional identities. It's a "through the looking glass" moment, where once you've seen the potential, there's no going back.

Based on "Ten Myths About the U.S. Tax System (Update) | Freakonomics Radio" from Freakonomics Radio Network Watch the original video

Unmasking Washington's Fiscal Fantasies: A Deep Dive into America's Tax and Debt Illusions

As spring blooms and the nation's thoughts turn to two enduring pastimes—baseball and taxes—it's clear that both are undergoing significant rule changes. While Major League Baseball embraces technology to refine its calls, the U.S. tax system and its accompanying fiscal challenges are grappling with rule changes and, more importantly, pervasive myths.

Enter Jessica RLE, a budget and tax fellow at the Brookings Institution, and a veteran of Washington's often-unseen policy battles. Consistently named one of D.C.'s most influential economic policy professionals, RLE is a self-proclaimed "friendless soul" whose non-partisan approach involves being "critical of everybody in Washington." Her career, marked by "sharing uncomfortable truths and frankly angering people," has been dedicated to sounding the alarm on two critical issues: an escalating federal debt crisis that is "even worse than you think," and the widespread misunderstanding of U.S. tax policy.

RLE's journey into the labyrinth of federal finance began in high school, sparked by a US News & World Report challenge to readers to balance the budget. "I just rolled up my sleeves given the nerd I am and went, 'This is going to be fun,'" she recalls. This early fascination matured into a deep understanding of how taxes and spending illuminate "the philosophical questions of what is the role of government." Her quarter-century in Washington has included stints at the Heritage Foundation, six years as chief economist to Senator Rob Portman, and work on several presidential campaigns, including for Marco Rubio and Mitt Romney.

Despite her deep involvement in partisan politics, RLE identifies as "pragmatic and right of center," advocating for free markets and less spending, but firmly independent. Her work is focused on being an "honest broker," calling out errors from both sides of the aisle. She criticizes the Biden administration for its "too big government and too big spending" approach, which she argues added $4 trillion in new spending and overheated the economy. Similarly, she points to the Trump administration for adding $8 trillion in new spending and tax cuts in its first term, and for pursuing "aggressively inflationary" policies like tax cuts, increased spending, tariffs, and immigration crackdowns. "I pull my hair out most days because I see two sides that are Dunning-Kruegering up and down, screaming at each other when both are making big mistakes," she laments, referring to the cognitive bias where those with the least knowledge are often the most confident.

The Great Deception: Unpacking the Top 10 Tax Myths

RLE's recent article, "Correcting the Top 10 Tax Myths," serves as a vital guide for navigating the often-misleading discourse around federal taxation. Written "with my hair on fire and smoke coming out of my ears" due to the prevalence of misinformation, the piece aims to equip the public with accurate information for a more informed national debate, especially as Washington prepares for consequential tax policy decisions.

She argues that "false narratives about taxes" pervade both conservative and liberal ideologies. Conservatives, she notes, "vastly overrate the positives of tax cuts," clinging to the "magical power" myths that tax cuts pay for themselves or force spending cuts. Liberals, conversely, champion an "equity distribution narrative," claiming the middle class bears the brunt of taxes while the wealthy and corporations pay nothing, and that taxing the rich can effortlessly solve deficits. RLE asserts that both sides are "promising their voters a free lunch."

Let's dismantle these pervasive myths, one by one:

  1. Myth 1: Tax Cuts Pay for Themselves.

    • Reality: While tax cuts can stimulate some economic activity and generate additional revenue, they "almost never pay for themselves." History shows they typically lead to increased deficits.

  2. Myth 2: Tax Cuts Will Starve the Beast.

    • Reality: The idea that cutting taxes will force Congress to reduce spending is historically false. RLE's research indicates the opposite: when taxes are cut, Congress tends to increase spending. Conversely, when taxes are raised, spending often decreases.

  3. Myth 3: The Middle Class Pays Higher Tax Rates Than the Rich.

    • Reality: This is demonstrably untrue. When all combined federal taxes are considered, the top 1% of earners pay an average rate of 33%, the middle class pays around 12%, and the bottom earners pay roughly zero. The U.S. actually has the most progressive tax system in the OECD.

  4. Myth 4: The 91% Tax Rates of the 1950s Produced Massive Revenue.

    • Reality: While the statutory rate was high, virtually nobody actually paid the 91% rate, or even over 50%. These high brackets raised "virtually no revenue" due to numerous deductions and loopholes.

  5. Myth 5: Europe Funds Its Bigger Governments by Taxing the Rich More.

    • Reality: Europe taxes its wealthy at roughly the same rate as the United States. The significant difference in overall tax revenue comes from Value Added Taxes (VATs), which are essentially national sales taxes that disproportionately impact the middle class.

  6. Myth 6: Tax Cuts for the Rich Are the Primary Reason for Large Budget Deficits.

    • Reality: While tax cuts contribute, spending is a much larger driver. Since 2000, tax cuts have accounted for about 2% of GDP, with only about 0.6% of GDP attributable to cuts for the rich. During the same period, spending has increased by 6% of GDP.

  7. Myth 7: Taxing Corporations and Millionaires Can Eliminate the Deficit.

    • Reality: Even if you were to tax corporations and millionaires at 100% and seize all their wealth, it "doesn't come close" to covering the federal deficit. The numbers simply don't add up.

  8. Myth 8: Most of the 2017 Tax Cuts Went to Corporations and the Wealthy.

    • Reality: While the wealthy and corporations received larger cuts in pure dollar terms, as a share of the taxes they were paying, it was a "roughly proportional income tax cut." Everyone saw their tax rates drop by about one percentage point.

  9. Myth 9: Reverting to the 1980 Tax Code (Repealing Reagan, Bush, and Trump Cuts) Would Offer Painless Deficit Reduction.

    • Reality: Such a move would send the tax burden on the middle class "through the roof" to "unacceptably high levels," not just affecting the rich.

  10. Myth 10: America's Corporate Taxes Are Far Below International Standards.

    • Reality: The U.S. had the highest corporate tax rate in the developed world until 2017. Even after the Trump-era cuts, our statutory and effective corporate tax rate remains in the top third globally. When pass-through corporations are included, the U.S. collects slightly more in business taxes than other countries.

RLE also addresses common misunderstandings, like the famous Warren Buffett quote about his secretary paying a higher tax rate. She explains that much of Buffett's income comes from capital gains, which are only taxed upon sale. However, even accounting for capital gains, high earners still pay significantly higher rates overall. She criticizes the Biden administration's calculation of rich people's tax rates as "pretty dishonest," arguing they inflated wealth figures (including unrealized gains) while undercounting corporate and estate taxes paid by the wealthy.

This intellectual dishonesty, RLE suggests, is a product of Washington's "sedition of power." Economists and policymakers, even those with intellectual integrity, can become "hacks" when they enter positions of influence, believing in a "noble lie" that justifies playing dirty to achieve desired policy outcomes. RLE, however, maintains that her "credibility in Washington is all you have," making such compromises untenable for her.

The Unstoppable Avalanche: America's National Debt Crisis

The widespread misunderstanding and manipulation of tax policy have dire consequences. RLE points to a "disaster of a tax code" that is "extraordinarily complicated," "inefficient," and "doesn't raise enough money to fund our spending." Since 2000, about one-third of the rise in deficits can be attributed to tax policy, while two-thirds stem from spending policy.

The federal deficit, the annual gap between spending and revenue, hit $1.8 trillion last year. Stacked year after year, this accumulates into the national debt, which currently stands at around $39 trillion—a staggering 124% of GDP, the highest since World War II.

The cost of this debt is rapidly becoming unbearable. Interest payments alone have tripled since 2021, from $350 billion to nearly a trillion dollars annually, projected to reach $2 trillion within a decade. This makes interest the second-largest item in the federal budget, surpassing Medicaid, defense, and Medicare, and on track to overtake Social Security by 2042.

RLE identifies the "main villain" as a federal government that "absolutely cannot stop spending money." After a period of fiscal responsibility from 1985 to 2000, lawmakers "threw out any sort of fiscal responsibility" after the budget was briefly balanced. An "arms race" has emerged where neither party believes it can win elections without promising "big tax cuts and big spending increases for everybody."

"Politicians don't know how to win otherwise," RLE states, describing Washington as a "flock of tweens who just discovered Klarna or Afterpay and they just go crazy buying every pair of shoes and gaming system." She recounts strategy sessions where lawmakers privately acknowledge the unsustainability of the debt but refuse to speak publicly, fearing electoral defeat. "I'm just going to try to pander the best I can and hope that when the consequences come, my successor is in office instead of me."

The Untouchable Solutions: Entitlement Reform and Middle-Class Sacrifices

To address this looming catastrophe, RLE highlights two particularly "vexing" and politically "third rail" solutions: entitlement reform and the need to consider broader tax increases.

Entitlement Reform: The most significant drivers of future deficits are Social Security and Medicare. A common myth is that these programs are fully funded by payroll taxes and cannot run deficits.

RLE's proposed solutions for Social Security involve three levers: raising taxes, raising the retirement age, and reforming benefits. Her plan includes new taxes, a higher age, and—most controversially—lower benefits for high earners. Means-testing, where benefits are reduced for wealthier individuals, is a logical starting point, especially since Social Security was designed as a "poverty prevention program," not a "universal huge get-rich benefit." However, means-testing remains a "third rail" in Washington, as politicians fear electoral suicide, and many voters mistakenly believe their benefits are fully pre-funded savings.

RLE posits that even a powerful figure like a sitting president publicly surrendering their own Social Security benefits to lead by example might not overcome partisan division. While a Republican president like Donald Trump might rally his party to support reform, Democrats would likely view it as a partisan attack aimed at cutting Social Security to fund tax cuts for the rich.

The Need for Broader Revenue: While not extensively detailed in the transcript, the implication throughout RLE's analysis is clear: the scale of the debt is so immense that even optimal tax policy, which might raise revenues from the current 17% of GDP to 20%, "is still not going to be enough to fix the budget" because spending is projected to rise to 33% of GDP. This underscores the uncomfortable truth that solving the fiscal crisis will require not only significant spending cuts, particularly in entitlements, but also potentially broader revenue increases that extend beyond just the wealthiest Americans.

A Path Forward?

The challenge, RLE concludes, is how to convince "the inmates to lock the asylum." Politicians pander to voters with promises of big tax cuts and big spending, deferring the costs to future generations. The only viable path, she believes, is for both parties to "privately come together and put everything on the table," encompassing tax reform, Medicare, and Social Security. Everyone must be "working together and everybody is sacrificing."

However, based on the current political landscape, RLE offers a grim outlook. The persistent reliance on "outdated, simplistic, and false assumptions about the federal tax system" by both parties continues to produce "destructive tax policies" and an ever-deepening fiscal hole. Until politicians find the courage to prioritize national solvency over electoral gain, America's fiscal future remains clouded by the very myths that perpetuate its decline.

Based on "AI Agents & Claude Skills Full Course: Setup and Build" from Greg Isenberg Watch the original video

The Unsexy Secret to Supercharging Your AI Agent's Productivity

Forget the flashy templates and pre-built agents. True AI mastery lies in teaching your digital assistant like a new employee, one custom "skill" at a time.

In an era where AI agents promise to revolutionize work, many users find themselves frustrated, wondering why these seemingly intelligent systems aren't living up to the hype. The problem, according to AI expert Ross Mike, isn't the models themselves. "The models are exceptionally good," he asserts, pointing to the impressive capabilities of advancements like Opus 4.6 and GPT 5.4. The real challenge, he contends, lies in how we guide them.

"We've reached a point where the models are good," Ross Mike explains, "but context still matters, and you have the power to steer the models in a direction where you can get quality or you can get slop." His message is clear: if you want to unlock peak productivity from your AI agents, you need to understand and master the art of context management and custom skill building. And often, that means doing the unsexy, iterative work yourself.

The Context Conundrum: Why Less is Often More

At the heart of an AI agent's operation is its "context window" – the information it assembles to execute an action. This includes general system prompts provided by the model developer, built-in tools (like read/write functions for code), the current codebase (if applicable), and, crucially, the ongoing user conversation.

A common misconception, Ross Mike notes, revolves around agent.mmd or cloud.mmd files. Many users believe these extensive files, packed with instructions and background information, are essential for guiding their agents. However, he argues that 95% of the time, they're not just unnecessary, but detrimental.

"Imagine I told you, Greg, every time we're about to shoot a podcast, 'Greg, you need a microphone.' You know you need a microphone, right? You've done this plenty of times," Ross Mike illustrates, highlighting the models' inherent capabilities. If an agent has a codebase in context, it doesn't need to be explicitly told it's using React; it can infer that from the code itself.

The major drawback of agent.mmd files is that their entire content is added to the agent's context in every single turn of a conversation. A thousand-line agent.mmd file could consume 7,000 tokens per run, leading to significant costs and, more importantly, a quickly saturated context window.

The Power of Progressive Disclosure: Embracing "Skills"

This is where "skills" come in – a more efficient and intelligent way to provide agents with specialized knowledge. Unlike agent.mmd files, skills leverage a mechanism called "progressive disclosure." When you create a skill (e.g., a skill.md file), only its name and description are initially added to the agent's context. The bulk of the information—the "bunch of info" as Ross Mike puts it—is only loaded when the agent determines it needs that specific skill.

"We're talking thousands of tokens compared to a couple hundred," he explains. This method not only saves tokens but also keeps the agent's context window leaner, allowing for better performance. A less cluttered context window prevents the agent from getting "dumb" as the conversation progresses, much like how a human struggles to process information when overwhelmed.

How to Build the Perfect Skill: Treat Your AI Like a New Hire

Ross Mike's most surprising and impactful advice is on how to build these skills. The biggest mistake, he says, is identifying a workflow and immediately trying to create a skill for it. Instead, he advocates for an iterative, hands-on teaching approach, akin to mentoring a new employee.

"You would ideally like them to fail and then you want to then tell them, 'No, this is how you do it.' Like there needs to be some sort of experiential learning," he says, drawing a parallel to human training.

He shares an anecdote from his own YouTube channel, where he uses an AI agent to vet potential sponsors. Initially, he simply told the agent: "Check every 15 minutes for emails, do research on a sponsor, and tell me if they're worth it." The result? Every sponsor was deemed "legit," with no critical analysis.

This led to a crucial realization: "The model needs a step-by-step guide." Instead of writing a skill immediately, Ross Mike walked the agent through the process:

  1. "Okay, I just sent you an email. Tell me about the company."
  2. "Check their Twitter, check their YouTube, check their Trustpilot, check if they've raised any money."
  3. "If two of these don't exist or are not in good standing, automatic rejection."

Only after successfully guiding the agent through the complete workflow, witnessing it make correct decisions and perform the required actions, did he then tell the AI: "Review what you did and then create the skill."

This "recursive skill building" doesn't stop once a skill is created. AI agents will still encounter edge cases and failures. When a skill-driven task fails, Ross Mike advises:

  1. Don't complain; thank God! This is an opportunity to improve.
  2. Identify the error: Ask the agent why it failed. It will often provide a descriptive error.
  3. Guide the fix: Tell the agent to fix the identified problem.
  4. Update the skill: Once the agent successfully executes the fix, instruct it to update the skill file so the error doesn't recur.

This continuous feedback loop makes skills incredibly robust and tailored to your specific needs. Ross Mike's YouTube report generator, which flawlessly pulls data from eight sources, is a testament to this method, having gone through five iterations of recursive skill building.

Scale for Productivity, Not Looks

This philosophy extends to how you structure your AI agent ecosystem. Ross Mike cautions against the temptation to immediately set up multiple sub-agents and download dozens of pre-made skills from marketplaces.

"I don't download skills because your agent needs the context of a successful run, which you then turn to skills," he states. Beyond the lack of tailored context, downloading third-party skills also poses significant security risks.

Instead, he advocates for a gradual, purpose-driven expansion:

  1. Start with one main agent: Build up its core skills for your primary workflows.
  2. Introduce sub-agents strategically: Once workflows are predefined and robust, create sub-agents to manage specific domains (e.g., a marketing sub-agent). These sub-agents, too, should have their own context and skills, built through the same iterative process.

"It's not sexy," Ross Mike admits, acknowledging that this hands-on approach requires effort. "But you sort of have to put in the work and build it up."

The Future is Personal: Your Workflow is Your Edge

As AI models continue to improve, Ross Mike believes the "harness" – the tools, context, and strategies we wrap around them – will become even more critical. Less is more when it comes to context, and what truly matters is codifying your unique workflow, your specific taste, your specific strategy into skills.

"The one thing that you and I have that the models don't have is my specific workflow, my specific taste, my specific strategy of doing things," he emphasizes. This is why building your own skills, based on your own successful interactions, is invaluable.

For developers, even the tech stack details become less relevant in agent.mmd files. Code itself now serves as context. A solid template for a web or mobile app can provide the initial context an agent needs to build upon, negating the need for verbose, framework-specific instructions.

In a world increasingly shaped by AI, understanding how to effectively build and guide these agents isn't just about productivity; it's about staying relevant. As Ross Mike puts it, "I genuinely believe anyone who knows how these tools work and like knows how to build agents and like craft skills and like knows how to make them productive, we're in a for a good run."

The path to AI mastery isn't about finding the perfect pre-built solution. It's about getting your hands dirty, teaching your AI like a diligent mentor, and recursively refining its "skills" until it perfectly mirrors your unique approach. In doing so, you're not just creating a productive agent; you're future-proofing your own capabilities in the age of artificial intelligence.

Based on "I built a custom Slack inbox. It was easier than you think. | Yash Tekriwal (Clay)" from How I AI Watch the original video

Taming the Digital Deluge: How One Hyper-Optimizer Built a Custom Slack Inbox with AI

In the relentless tide of modern work, few things generate more low-grade anxiety than a burgeoning Slack inbox. For Yash Tekriwal, Head of Education at Clay, this wasn't just an occasional annoyance; it was a daily deluge. Waking up to 100 to 150 new Slack notifications, often with 60-80% falling into the "for your information" (FYI) category, meant sifting through a mountain of noise to find the truly actionable 30-40 messages. This wasn't just inefficient; it was mentally draining.

"My 100 to 150 that's giving me anxiety is actually more like 30 to 40 that I really need to be on top of," Tekriwal explains. This stark reality drove him to envision a better world – a personalized digital assistant that could not only categorize and summarize his communications but also build a custom tool to manage them. The result? A bespoke Slack inbox, built with AI, that transforms chaos into a clear, actionable workflow.

The Problem: Drowning in Notifications

Tekriwal’s struggle is a common one. Slack, an indispensable communication tool, often suffers from a "notification parity" problem. A direct message about an urgent project update receives the same visual cue as a colleague's amusing photo in the "fun dog channel." This lack of inherent prioritization means every notification feels equally important, leading to a constant state of alert and the fear of missing something critical.

"Not all notifications are created equal in Slack," Tekriwal notes, highlighting the core issue. His ideal solution involved not just filtering unread messages, but intelligently categorizing them by type (DMs, group DMs, threads, @mentions) and, crucially, by required action:

This vision, however, required more than just a new habit; it demanded a custom-built solution.

The AI-Powered Journey: From Proto-Code to Polished App

Tekriwal's journey to his custom inbox began with a clear understanding of AI's dual power: "You can use AI to do a task for you like categorize things, summarize things, or you can use AI just to build a tool that would have been much harder to build before with very straightforward APIs and structured data."

Phase 1: The "Jarvis Digest" with OpenClaw

His initial foray involved an AI coding agent named OpenClaw, operating within Discord – chosen for its superior threading and search capabilities compared to Telegram. Tekriwal engaged in a lengthy, iterative dialogue with OpenClaw, essentially reverse-engineering Slack's intricate notification system. The goal was to programmatically pull only relevant messages with specific context, based on timestamps and read status.

The output was a "Jarvis digest channel" within Slack itself. This channel provided a text-based summary of his notifications, grouped into his desired categories: direct @mentions, DMs, group mentions, and threads, each with sub-categorizations of action, read, or FYI. While a significant improvement, consolidating messages into a single, scrollable text digest still presented a cognitive load. "I had to scroll at least four to five screens down just to be able to even see all of the notifications that are already summarized for me," he recounts. This proved draining, prompting the need for a more intuitive interface.

Phase 2: Building the Dream UI with Perplexity Computer

This is where Perplexity Computer entered the scene, offering a leap in capability. Tekriwal's vision was a clean, navigable software interface – a "Superhuman for Slack" dashboard. Perplexity Computer delivered.

What made Perplexity Computer the ideal tool for this next phase?

The result is a visually intuitive, Kanban-style dashboard. It features three distinct columns:

The dashboard allows for customizable grouping (e.g., DMs first) and, most crucially, a "magic" Archive All button for the FYI column. Clicking this not only clears them from the dashboard but also marks them as read in Slack, providing a sense of control and calm that native Slack lacks. "This is such a better way to just get through your queue," observes Claire Vo, host of "How I AI."

Beyond Slack: A New Era for Software and Productivity

Tekriwal's custom Slack inbox isn't just a personal triumph; it's a powerful demonstration of AI's transformative potential for software development and productivity.

Is SaaS Dead? Not Exactly.

The rise of AI-powered custom tools sparks a debate: Is this the end of traditional Software-as-a-Service? Tekriwal argues the opposite. "Slack is still great," he contends. "It's great for sending messages... but to get it to 10 out of 10, I'm just going to build the thing that works with my brain." AI doesn't kill SaaS; it creates "SaaS Custom," allowing individuals to build personalized layers on top of existing platforms, bridging the gap between a generic tool and an ideal individual workflow.

The "Cambrian Explosion" of Niche Software

The most profound implication, according to Tekriwal, is an "explosion in software being created and used." AI dramatically lowers the cost of building, making it feasible to create highly specific applications that might never have justified venture funding or the time of traditional developers.

"My dream is for someone else to watch this video and say, 'I want to build that app on top of Slack,'" he shares. "And then I can go pay that person $15 a month for this app to be maintained and used... because I would happily pay them." This vision points to a future where a "Cambrian explosion" of niche businesses caters to specific needs, creating a vibrant ecosystem of useful, bespoke software.

The "Anti-To-Do List" Philosophy

Tekriwal also champions the "anti-to-do list" – a powerful framework for leveraging AI. Instead of listing tasks to complete, list tasks you never want to do again. This includes mundane, repetitive chores like manually deleting spam, entering meeting action items, or, of course, sifting through unprioritized Slack messages. Spending an hour a day automating these anti-to-dos with AI, he argues, is an incredibly worthy investment of time.

AI for Team Collaboration and Prototyping

Perplexity Computer's utility extends beyond personal productivity. Tekriwal's teammate, Chris Ming, used it to prototype a persona-based learning journey for Clay University's website. By allowing Perplexity Computer to access the existing site in the browser, it could visually recognize the design and help build a new, tailored UI. This capability to "build a visual bridge" between design and other stakeholders proves invaluable in accelerating development and improving cross-functional communication.

The Human Touch: AI for Play (and Serious Fun)

While Tekriwal primarily views AI as a work tool, even this "hyper-optimizer" finds joy in its capabilities for personal pursuits. He leverages AI for brainstorming and research, particularly for planning elaborate social events. For his annual "Olympics" with friends, AI helps generate unique game ideas (like "sword, fish, or soup" trivia) and even orchestrate complex team rotations to ensure everyone interacts. He's also used it to discover new board games, proving that even for the most pragmatic users, AI can enhance the fun.

Tekriwal's custom Slack inbox is more than just a clever hack; it's a blueprint for a new paradigm of digital living. It demonstrates how AI empowers individuals to reclaim control over their digital environments, transforming overwhelming noise into actionable insights, and paving the way for a future where software is not just universally functional but deeply personal. The tools are here; the only limit is imagination.

Based on "The Era of AI Agents | Aaron Levie on The a16z Show" from a16z Watch the original video

The Agent Uprising: How AI is Reshaping Software, Work, and the Enterprise

The future of technology, once envisioned as sleek interfaces designed for human interaction, is rapidly pivoting towards a new, more autonomous paradigm: AI agents. These intelligent entities are not just tools; they are becoming users, collaborators, and even decision-makers within our digital ecosystems. This profound shift, as discussed by Box CEO Aaron Levie on The a16z Show, promises unprecedented leverage but also introduces a complex web of challenges for individuals, startups, and established enterprises alike.

Levie contends that the "diffusion of AI capability is going to take longer than people in Silicon Valley realize," primarily because the implications stretch far beyond mere technological advancement. It demands a fundamental re-evaluation of how software is built, how work is done, and how trust and security are managed in an agent-driven world.

Software for the Non-Human User

The core thesis is startlingly simple yet revolutionary: "If you have a hundred or a thousand times more agents than people, then your software has to be built for agents." This isn't just about making APIs accessible; it's about fundamentally redesigning systems for non-human interaction. We are moving towards a world where our tools need an "agent interface" as much as, if not more than, a human interface.

The most effective paradigm emerging, according to Levie, involves "coding agents" – AI entities capable of accessing SaaS tools, knowledge workflows, and context, then "coding its way or uses APIs" to achieve tasks. This isn't just about data analysis; it's about active problem-solving and execution within existing software environments, a "superpower" that is already starting to compound.

The Algorithmic Divide: Humans vs. Agents

While the potential of agents is immense, a significant hurdle lies in the human element. "Algorithmic thinking is really really really hard for the vast majority of people who have jobs," Levie observes. He illustrates this by noting that most people would struggle to create a flowchart for their daily tasks, let alone explain complex processes to an AI agent. In a 50-person marketing team, perhaps only one person could truly document the intricate workflows.

This "impedance mismatch" means that for agents to be truly effective, humans must learn to articulate their jobs as systems. Levie draws an analogy to the advent of spreadsheets: early managers hired interns to do the "spreadsheeting," but soon, those managers themselves became proficient. The abstraction layer moved up. Similarly, the "rocket science" of orchestrating 42 agents today will "evaporate in very short order," becoming a standard skill set. The job simply "moves up a rung," demanding a new set of skills focused on system thinking and agent orchestration.

The Anthropic growth marketer, a viral example of one person automating the work of five to ten, exemplifies this shift. It shows the leverage possible when a systems-thinker is paired with an "infinite pool of engineers" (the agent) to automate workflows.

The Evolution of Agent Interaction: Beyond Code

Interestingly, Levie suggests that the way agents interact with software is also evolving. While initial excitement focused on agents "generating code," he now sees a different trajectory: "We started with code, then we went to the terminal, which is actually less code. And now this year is going to be the year of computer use." Agents are becoming "much more like humans using computers than them generating code."

However, this distinction might be semantic. Levie's own company, Box, is developing agents that can determine whether to use an existing skill, an existing tool, or "write code to solve that problem" on the fly. This flexibility, where an agent can fluidly switch between leveraging pre-built functionalities and custom coding, is "incredibly useful" for handling unforeseen or highly specific operations. The goal isn't just to replace human code generation but to empower agents to navigate the vast, often underutilized, capabilities of existing software.

The Enterprise Conundrum: Security, Trust, and Rogue Agents

While personal productivity with agents might involve giving them their "own phone number, own credit card, own Gmail account" – essentially treating them like separate digital humans – the enterprise context is vastly more complex. The idea of 50 humans and 50 agents collaborating in a shared space, with agents potentially gaining unauthorized access or causing unintended actions, sends "CFOs, CIOs, etc. running around trying to with their hair on fire."

The fear isn't just about practical integration; it's about control and liability. As one co-host put it, "you put people creating new integrations and you just say please break my system of record." The risk is "a thousand times greater" with agents, which can "leak your information whenever they want" if social engineered. The "context window" problem – ensuring an agent keeps sensitive information secret – remains a "very hard problem to solve."

Levie envisions a "read-only version of this for a number of years" where agents primarily consume information. The ability to "log in as them" and undo mistakes, while retaining oversight, clashes with the idea of agents operating as independent entities. This leads to a fundamental breakdown: agents cannot be treated exactly like humans because they lack privacy and carry immense liability for their human operators. This tension highlights why the "diffusion of AI capability is going to take longer than people in Silicon Valley realize."

The Great Divide: Startups vs. Enterprises

This risk aversion creates a significant chasm between startups and established enterprises. Startups, with "nothing to blow up," can embrace agents from the ground up, moving "much much faster." Enterprises, like JP Morgan, face an immense challenge integrating agents into their complex, legacy systems without compromising security or stability.

This divide will also impact the business models of existing SaaS vendors. Historically, they've sold "intelligence and domain expertise" packaged within their UIs. Agents, however, "want to only buy the data now" and have "unlimited access to the data," which has never been the core business model for many. This tension, previously seen with API access debates at companies like Salesforce, Workday, and SAP, is now amplified.

Levie believes that while "it's just absurd to think you're going to vibe code your way to like SAP," these legacy systems will face immense pressure. Much of their "domain knowledge" isn't just in a "well-orchestrated data layer" but embedded in UIs, middleware, and usage patterns. This complexity will slow the agentification of these critical data sources.

"Build Something Agents Want": The Meritocracy of Technology

Ultimately, Levie champions the idea of "build something agents want." At some point, agents will largely be in charge of choosing the tools they implement and use. This isn't about marketing to agents with flashy interfaces; it's about fundamental system quality. In a "breaking podcast news" moment, Levie refutes the notion that agents prioritize good APIs or IDLs. Instead, he argues, "it's the semantics that end up mattering a lot more... they actually have the collective wisdom of our experience using these platforms."

Agents, he suggests, are "pretty smart at choosing the right technology" based on "cost parameters... durability... meaningful stuff, not interface stuff." This will push the industry to build "better systems." If your tool is closed off or performs poorly for an agent, it will "eventually find a better tool for that company to go use." This could lead to agents dictating IT stacks, telling companies, "You need to finally rip out your legacy HR system or I'm not going to be able to automate this workflow for you."

This future, where "a hundred or a thousand times more agent volume on software than people" drives software evolution, means businesses will correlate their performance to "how well your agents can get access to the information they need." The new problem for software companies becomes: can you build "really really high quality APIs," monetize them effectively, and handle agent identities and access controls?

The era of AI agents is not just a technological upgrade; it's a paradigm shift that will redefine the relationship between humans, software, and intelligence. It promises immense leverage and efficiency but demands a thoughtful, cautious approach to security, integration, and the very nature of work itself. The journey will be long and complex, but the destination is a world where software is truly built for the intelligent machines that increasingly power our lives.

Based on "A Cease-Fire in Iran" from New York Times Podcasts Watch the original video

High Stakes, Fragile Peace: Inside the Last-Minute US-Iran Ceasefire

Just hours before a Trump-imposed deadline threatened a catastrophic escalation, the United States and Iran announced a temporary ceasefire. But as White House correspondent David Sanger reveals, the path to this precarious pause was paved with unprecedented threats, and the deal itself is fraught with discrepancies and enduring consequences.

The news broke like a sudden calm after a raging storm. "Fox News alert. We have a deal for now." From the New York Times, Rachel Abrams reported on a development that had the world holding its breath: the United States and Iran had agreed to a two-week pause in hostilities. Just hours before an 8:00 p.m. deadline that could have triggered a "massive escalation" in the war, both nations confirmed a temporary ceasefire, with Iran's state-run media instructing all military units to "stop firing" in accordance with the Supreme Leader's orders.

But the relief was tempered by immediate questions. What exactly had been agreed upon? And how did the world arrive at the precipice of a full-scale war, only to step back at the very last moment? To unravel the complexities of this dramatic turn, chief White House correspondent David Sanger provided crucial insights into the last-minute deal and the arduous path ahead.

A Deal in Discrepancy: Two Visions of Peace

While President Trump declared a 14-day ceasefire during which Iran would commit to the "full reopening" of the Strait of Hormuz, allowing the critical flow of oil, fertilizer, and even helium for semiconductor production to resume, the Iranian account painted a different picture. Iran's foreign minister, Abbas Aragi, a seasoned negotiator, stated that Iran would "cease their defensive operation for a period of two weeks" but crucially added that "safe passage through the Strait of Hormuz would only be possible by coordinating with Iran's armed forces."

This "daylight between these two statements," as Sanger noted, was significant. Trump's vision implied a return to the pre-war status quo, while Aragi's suggested Iran would retain the control it had seized over the past five or six weeks, dictating the pace and potentially even charging tolls for passage. The American bombing of Iran did stop almost immediately, but the fundamental disagreement over the Strait's control underscored the fragility of the agreement.

Even Israel, a major actor in the conflict alongside the United States, offered only lukewarm support. While the White House claimed Israel had agreed, a statement from the Israeli prime minister's office merely said they "support President Trump's decision," with Sanger observing they weren't "terribly enthused about it."

The Road to the Brink: A Cascade of Threats

The path to this temporary ceasefire was marked by an extraordinary escalation of rhetoric, primarily from President Trump. The initial impetus for the conflict had been the closing of the Strait of Hormuz, which was exerting immense pressure on the global economy. Trump's initial demands for Iran to reopen the Strait, first with a 48-hour ultimatum, then a 10-day negotiation clock, were largely ignored or mocked by the Iranians on social media.

The tension was briefly interrupted by a sudden crisis: the shootdown of an F-15E fighter jet on a Friday, leading to a daring and costly rescue operation for a missing weapons officer. This successful retrieval, as Sanger's colleague Eric Schmidt reported, seemed to embolden President Trump, who then resumed his threats with renewed vigor.

What followed was a series of "truly bizarre social media posts" that pushed diplomatic norms to their breaking point:

This final post, interpreted by many as a threat to "wipe out 90 million people" and commit a "huge war crime," sent shockwaves across the political spectrum. Democrats called for invoking the 25th Amendment, but criticism extended even into the MAGA movement, with figures like Alex Jones, Marjorie Taylor Green, Candace Owens, and Tucker Carlson expressing alarm. Senator Ron Johnson, typically reticent to criticize Trump, stated, "this simply is not how Americans speak and how presidents ought to speak. I do not want to see us start blowing up civilian infrastructure."

Despite the extreme rhetoric, evidence of military mobilization beyond existing operations was hard to discern. Yet, with significant American forces already in the Middle East, the capacity to fulfill these threats was undeniable.

A Perilous Pause: What Lies Ahead?

The sudden announcement of a ceasefire, brokered indirectly through Pakistani diplomats, brought a collective "sigh of relief." But as Sanger cautioned, "ceasefires are fragile by nature." The immediate test would be the resumption of traffic through the Strait of Hormuz, and whether it could return to pre-February 28th levels. Iran's newfound leverage over global commerce, however, is a power they are "not likely to give up anytime soon."

Beyond the Strait, the larger issues loom, particularly the nuclear weapons program. Trump's stance on Iran's nuclear material has been inconsistent, ranging from "cannot tolerate Iran having any nuclear material" to "I don't really care. It's buried so deep. We can watch it by satellite," before likely reverting to a demand for complete removal. This shifting position complicates negotiations, especially as a deal that leaves Iran with significant near-bomb-grade uranium would be seen as a weaker outcome than the 2015 Obama-era agreement.

While the "ratcheting up of pressure certainly played some role in forcing an agreement," terrifying the region and China, it's unclear if it genuinely "terrified the Iranians," given their complex decision-making process.

The Enduring Scars of Conflict

Even if this temporary ceasefire blossoms into a permanent peace, the United States and Israel, despite having "taken out in the opening hours of the war, the Supreme Leader," devastated military and intelligence assets, and set back missile programs, may find they have accomplished "virtually none of its major goals." Iran remains under the control of its current elite, still possesses nuclear material (with the potential to rebuild), and crucially, has gained "a sense that they were able to stand up to the United States and Israel."

The consequences of this conflict are profound and enduring:

As the two-week pause begins, the world watches to see if this fragile agreement can hold. Meanwhile, in a related development, American journalist Shelley KDson, abducted in Baghdad by an Iranian-aligned Iraqi militia, was freed in exchange for imprisoned militia members, highlighting the ongoing anxieties and complexities in a region deeply scarred by recent events. The ceasefire may offer a breather, but the underlying tensions, shattered trust, and geopolitical shifts promise a long and uncertain road ahead.