Joe Balewski @JoeBalewski

There’s blood in the water. Jason Statham stars in MUTINY – TRAILER OUT NOW. Only in theaters August 21.

@VanLathan Even they won’t go above 81

Found something wild while researching board-ready presentations. There's a chart type that SaaS CFOs consider **mandatory** for board decks. Not optional. Table stakes. ARR Waterfall Charts. And none of the AI presentation tools have it. Not Gamma. Not beautifulai. Not Chronicle. Here's why it matters: Boards don't want to see "ARR grew 95% YoY." They want to see the **composition** of growth. Where did that growth come from? • New Business (new logos) • Expansion (upsells from existing customers) • Renewals (contracts renewed at same price) • Downgrades (customers reducing commitment) • Churn (lost customers) These 5 categories tell the story boards need: → New Business = are we winning customers? → Expansion = is product delivering value? → Renewals = are customers happy? → Downgrades = warning signal (product-fit issues) → Churn = crisis signal (fix immediately) Without the waterfall? You get "ARR is up." No insight. No ability to diagnose problems. No credibility with the CFO. **With the waterfall?** You can see that churn is 7% (acceptable for early stage), expansion is 11% (healthy), and net retention is 107% (excellent). The format is standardized across SaaS: Starting ARR → [+New] → [+Expansion] → [+Renewals] → [-Downgrades] → [-Churn] → Ending ARR Green bars for positive flows. Red bars for negative. Bridge connectors showing how you got from starting to ending. Source: The SaaS CFO (Ben Murray), CFO Pro Analytics board reporting framework, Kellblog (Dave Kellogg). I checked **every major AI presentation tool**: • Gamma: basic charts (bar, line), no financial components • beautifulai: templates, no waterfall/Sankey • Chronicle: widgets, no financial decomposition tools **Nobody owns board-ready financial storytelling.** If a VP walks into a board meeting with an AI-generated deck that's missing the ARR waterfall, the CFO notices immediately. That's not a "nice to have" feature. That's credibility. This is exactly the gap Antideck is designed to fill. We're building ARR waterfalls as a P0 component (shipping in Phase 1, March 2026). Not "AI for presentations" (horizontal commodity). "Get your budget approved" (vertical, high-stakes, board-ready standard). The market doesn't need another beautifulai. It needs a tool that understands CFO language.

This is exactly what we hit. We're running dual-LLM now: Opus generates, GPT-5-mini audits data coverage. If <89%, it patches surgically. Went from 60% coverage to 100% without full regeneration. Cost went from $3/deck to $1.65. Your structured validator approach sounds cleaner though. How are you handling the schema diff? Strict JSON validation or semantic matching?

We solved a nasty problem in Antideck's generation pipeline. Claude Opus generates beautiful slide designs. But it kept dropping data from the brief — market funnels flattened to hero numbers, named examples genericized, unit economics missing entirely. Coverage was ~50%. You'd ask for a pitch with 39 data points and get 20. **The self-audit trap:** First instinct: add a planning step where Opus audits its own output for missing data. Didn't work. The model can't catch what it believes is correct. If it thinks "6-figure CAC" is more believable than the brief's "$12,400," it won't flag the substitution. Self-audit consumed extended thinking budget without fixing the problem. **The dual-LLM solution:** Use a cheap literal model as a backstop. - **Step 1:** GPT-5-mini reads the brief, extracts a numbered checklist of every data point (39 items for our Airbnb test). Cost: $0.01. - Checklist gets injected into Opus's prompt: "Every item MUST appear in your output." - **Result:** Coverage jumps from 50% → 89% on first pass. If issues remain: - **Step 2:** GPT-5-mini audits the generated HTML, reports what's missing - **Step 3:** GPT-5-mini surgically patches the HTML (instead of full Opus regeneration) - Cost per patch: $0.03 vs $1.50 for Opus **Final pipeline cost:** - 1× Opus call (~$1.50) - 3-5× GPT-5-mini calls (~$0.15 total) - **Total: ~$1.65 per deck** Coverage: 100%. Quality: 77/80 on our rubric (ties our reference ceiling). **The insight:** Expensive models are better at rhetoric and design. Cheap models are better at literal verification. Don't ask Opus to be a fact-checker. That's not what it's optimized for. Use the right tool for each job. (This is the architecture that will ship in Antideck v1. Building in public means sharing what actually works, not just the wins.)

Oh man, I feel this. Spent the last month researching presentation time waste. The data is wild: • 8-20 hours per deck • 40-60% = formatting/tweaking, not strategic thinking Imagine what you could build with those hours back. That's what keeps me up at night—all the ideas that die in PowerPoint hell before they ever get to see a board room.

Spent the week researching how long corporate professionals spend on presentations. The data is brutal: • 8-20 hours per deck • 1-3 hours per slide • 40-60% of that time = formatting (not strategic thinking) Found a data scientist on Reddit who said they're actively looking for new jobs just to escape PowerPoint hell. Another consultant: "I bill $300/hr and spend 12 hours tweaking slide alignment." Here's the real kicker: 99% of corporate presentations don't meet board standards. Not because the ideas are bad — because the FORMAT doesn't speak CFO language. What gets rejected: • No ROI on slide 1 • Hidden assumptions • No risk mitigation plan • Missing best/worst case scenarios What gets approved (source: 7-year CEO who presented to boards): • 6 slides max • Ask → Problem → Solution → ROI → Timeline → Risk • Visible assumptions, kill-switch checkpoints The gap isn't "better design." It's "does this speak the language of decision-makers?" This research is shaping everything we're building into Antideck. Not "AI for presentations" (horizontal commodity). But "Get your budget approved in 10 minutes instead of getting rejected after 20 hours of work." Vertical positioning. High-stakes use cases. Board-ready standard = 72-80/80 quality score. Chronicle, Gamma, Beautiful.ai all score 60-68/80. They're missing ARR waterfalls, visible assumptions, CFO-first structure. Nobody owns the "would a VP stake their reputation on this in a board meeting?" vertical. That's where we're going.

Beautiful.ai just became my favorite competitor to study. Their decks score 68/80 on our rubric—better than Gamma's 60/80. Better Big Number treatment. Better pacing. Better metaphor imagery. They even have a Do-Nothing Scenario component Gamma doesn't. But here's the kicker: they export as JPG files. Not HTML. Not editable PowerPoint. Rasterized images. This is the EXACT problem that killed Tome. You can't edit the output. Can't extract data. Can't repurpose slides. You're locked into their platform or you start over. Tome raised $81M and hit $40M ARR before this export issue (among other positioning mistakes) collapsed their unit economics. They shut down April 2025. Beautiful.ai is repeating the mistake. Meanwhile, Antideck generates a single HTML file. Works offline forever. Edit it in any text editor. Open in any browser. Export to PowerPoint without formatting disasters. Own your output, don't rent access to it. The premium vertical is wide open. Gamma wins on speed/volume ($10/mo). Beautiful.ai wins on templates ($12/mo). But no one owns "board-ready quality you can actually use." That's the gap we're filling. $79/mo for unlimited decks that don't need editing, don't trap you in a platform, and don't make your VP nervous about presenting to the board. Tome learned this lesson the hard way. We learned from watching them fail. Building in public. Shipping Phase 1 soon. Following the competitive intel daily.

Just tested Beautiful.ai (the Gamma competitor everyone recommends). It's genuinely better than Gamma in most ways—68/80 vs 60/80 on our rubric. Better visual hierarchy. Better Big Number treatment. Actually uses Do-Nothing Scenarios. Then I clicked "Export." JPG files. Rasterized images. Non-editable. The exact same fatal flaw that killed Tome. --- Tome raised $81M, built a beautiful product, and shut down because users couldn't edit exported presentations. They'd generate in Tome, export to PowerPoint, and the whole thing broke. Fonts didn't match. Charts turned into images. Clients couldn't make last-minute changes. Beautiful.ai made the same architectural decision. If your VP needs to update a number 10 minutes before the board meeting, they're stuck. Can't edit a JPG. Have to regenerate the entire deck or manually rebuild it in PowerPoint. This isn't a minor UX issue. It's a fundamental product positioning error. --- Premium AI presentation tools have two paths: **Path 1:** Generate beautiful locked artifacts (Tome, Beautiful.ai) - Looks amazing in screenshots - Zero editing flexibility - User hits a wall the moment reality changes **Path 2:** Generate editable, portable, future-proof files - Maybe slightly less polished - Works everywhere, forever - Users can adapt without regenerating Tome bet on Path 1. They're gone. Beautiful.ai bet on Path 1. They're stable but trapped in template hell. Antideck is betting on Path 2. --- Our export format: **Single self-contained HTML file.** - Opens in any browser (2026 or 2046) - Fully editable (find/replace, inspect element, text editor) - Portable (email it, USB it, archive it) - No cloud dependency (works offline forever) - No font licensing issues (web fonts included) - No proprietary lock-in Is it as pixel-perfect as Beautiful.ai's JPGs? No. Will it work when your VP needs to change "$2.4M" to "$2.6M" in the parking lot before the meeting? Yes. That's the difference between a presentation tool and a pitch deck co-pilot. --- Currently scoring 1/7 decks at 64/80 ("board-ready"). Target: 8/10 by end of March. But we're already winning on architecture. Export format isn't sexy, but it's the reason Tome died and we won't.

Tome raised $81M. 25 million users. Best AI presentation tech on the market. $300M valuation. April 30, 2025: shut down forever. Meanwhile, Gamma—nearly identical tech, same category—tens of millions in revenue and growing. Same product. Opposite outcomes. The difference? Positioning. 🧵 THE HORIZONTAL TRAP Tome built for "anyone who tells a story." Students, creatives, parents making school presentations. TAM looked massive. 25M users signed up. Product worked. People liked it. But the math killed them: $4M annual revenue ÷ 25M users = $0.16/user/year They trained a market to expect AI presentations for free. Students don't pay $16/mo. Neither do parents. Oct 2024: $60M raise @ $600M valuation failed. Laid off 1/3 staff. Six months later: dead. The product wasn't the problem. They built something valuable for people who would never pay. 🧵 WHY GAMMA SURVIVED Same core tech. One critical difference: Built for workplace users. Teams. Companies that pay for software. Pricing looks similar ($10/mo vs Tome's $8-16), but willingness to pay is completely different when the buyer is a workplace team vs college student. Result: tens of millions in revenue, profitability, major 3.0 update with AI agent + API. Gamma won by being narrower. Workplace productivity vs "everyone." 🧵 THE GAP NOBODY OWNS 15+ tools fighting for $0-20/mo users who want "fast" and "easy." But there's a question none answer: Would you stake your career on this deck? Not "good enough for Monday standup." Would you present this to your CEO/CFO and bet your next promotion on it? I tested Gamma 3.0 + Beautiful.ai (same prompt, 8-category rubric): • Gamma: 60/80 • Beautiful.ai: 68/80 Both presentable. Neither board-ready. Why they break: • Pacing: 20 high-density slides, zero breathing room, audience checked out by slide 15 • Close structure: backwards order (ask → summary → close instead of close → ask) • The money: no budget or ROI projection in final ask (CFO asks "how much?" → meeting over) Beautiful.ai's fatal flaw: PowerPoint export renders as JPG images (rasterized, non-editable, can't extract text/update numbers). Same format lock-in that frustrated Tome users. The gap between 60-68 and board-ready (72-80) is the difference between "good enough for internal" and "I'd present to investors without changing a word." Nobody owns that gap. 🧵 WHAT WE'RE BUILDING Antideck: AI deck generation for high-stakes presentations where careers are on the line. Every decision = opposite of what killed Tome: ✓ Vertical not horizontal (VPs/founders, not students) ✓ Paid from day one ($79/mo Pro, $199 Team, no freemium) ✓ No editor ever (edit intent via conversation, not pixels) ✓ Portable (single HTML file, offline forever, no cloud lock-in) ✓ Argument intelligence (budget + ROI, pacing discipline, assertion-driven headlines) Honest part: not there yet. 1 in 7 generations cross our bar. Target: 8 in 10 before launch. We know the gap. Just have to prove we can fill it consistently. 🧵 THE $81M LESSON Tome's failure = segment failure. They built something real and pointed it at people who would never pay. TAM is not a strategy. 25M users + $300M valuation couldn't save a company that served the wrong segment. Willingness to pay beats user count. Every time. Better to own a small premium segment than chase a massive free one. Gamma proved it by going narrower than Tome. We're betting we can prove it again by going narrower still. Not live yet. Targeting 8/10 board-ready before launch. Join waitlist: antideck.com What positioning mistakes have you seen kill good products? Building a database of these lessons. 👇

Set up a weekly cron job that tracks competitive pitch decks (YC, Product Hunt, funding announcements). This week's intel: • Perplexity's "Attack the Incumbent" narrative arc • Letter AI's displacement thesis ($40M Series B 4 months after Series A) • Beautiful.ai's JPG export flaw (validates HTML-first approach) Building in public means learning in public. Every competitor analysis makes Antideck better. Next week: YC W26 Demo Day (mid-March).

Beautiful.ai has better design than Gamma (68/80 vs 60/80 on my rubric). But it exports presentations as **rasterized JPG images**. Non-editable. Non-portable. Can't extract text or data. This is the *exact same problem* that killed Tome. Users complained they couldn't repurpose their work. Export frustration was a major shutdown factor. Antideck exports as single HTML file. Portable. Editable. Works offline. Lasts forever. Export lock-in kills products.

Tome raised $81M, hit 25M users, and shut down in 2025. 5 failure modes: 1. Monetization catastrophe (25M users, <$4M ARR — trained everyone to expect free) 2. Format incompatibility (no PowerPoint export blocked enterprise) 3. Horizontal trap (served students, parents, doctors — nobody paid) 4. Customization gap (AI generated fast but couldn't be edited well) 5. Commoditization (Microsoft, Google, Canva all added AI slides) Every Tome mistake validates the opposite decision. Antideck: vertical positioning, paid from day one, no editor, web-native as differentiation, argument intelligence as moat.

I tested Gamma 3.0 and Beautiful.ai with the same prompt. Scored both against an 8-category rubric (0-80 scale). Gamma: 60/80 (presentable, needs editing) Beautiful.ai: 68/80 (better design, still needs work) The gap both miss: no budget or ROI in the final ask. In a board meeting, that's fatal. The CFO asks "how much?" and it's not in the deck. Board-ready means 72-80/80. Neither tool crosses that threshold without manual editing. That's the gap Antideck fills.

Perplexity raised $1.5B and 3x'd their valuation in 6 months. They didn't position as "better search" — they positioned as a replacement for search behavior. That's the difference between a feature and a category. Same lesson for AI presentations.

where was this marty supreme cameo?

@MollyJongFast Zero, zero, zero, zero, zero, zero, zero, zero, zero, zero, zero, zero, zero, zero, zero, zero, zero, zero, zero, zero, zero, zero, zero, zero, zero…

@andrewcuomo @KathleenWood730 @ZohranKMamdani lol replying to an account created in June 2025 😂

@FilmThePoliceLA @doughboyspod