Venu Javarappa

Adaptive Plugin: The next efficiency layer for Enterprise GenAI Enterprises are scaling LLM workloads at an insane pace - banking, healthcare, SaaS, telecom, government. But one silent issue keeps draining budgets: token waste. Oversized prompts. Long documents. Repeated context. Huge chat histories. All of that burns millions per year. We’re building Adaptive Plugin to solve this head-on. What Adaptive Plugin will deliver: Dynamic context resizing, semantic compression, and real-time prompt optimization that cuts 25–40% of tokens without losing meaning. Why enterprises are already showing early interest: 1. Significant cost reduction Large banks, insurers, legal teams, and SaaS companies spend six to seven figures annually on LLM tokens. Trimming a third of that is a direct budget win. 2. Lower latency without buying more GPUs Small, optimized prompts = faster responses. No new hardware, no larger models required. 3. Cleaner, more efficient AI Reducing tokens directly reduces compute load. It’s sustainability through pure efficiency - no greenwashing. 4. Built for the industries that suffer most from long context Finance: fraud systems, research, policy summaries Healthcare: EMR histories, clinical documentation Legal: contract review, case law analysis Tech/SaaS: copilots, agents, dev tools Telco & customer support: high-volume chat workloads Government: audit systems, citizen communication If your workload involves long documents or heavy chat threads, Adaptive Plugin hits immediate ROI. 5. Not a model gateway. Not prompt tweaking. Gateways optimize model choice. Manual engineering optimizes prompts. Adaptive Plugin optimizes tokens in real time, consistently, automatically. 6. Outcome-based pricing The model is simple: we earn only from the savings we generate. Where we are now We’re finalizing development and preparing for enterprise pilots. Full release is coming soon, and several major organizations are already in the early interest pipeline for 2026 deployments. We’re not launching today. But we’re close - and the efficiency gains are too big to ignore. At Friday AI, we’re building our first product: Adaptive Plugin, an efficiency layer designed to cut 25–40% of token usage in enterprise GenAI workloads through real-time context optimization and semantic compression. The product is fully developed and currently undergoing internal testing and early evaluations with selected partners. Our plan is simple: launch Adaptive Plugin in 2026 with a small group of enterprises, refine the product through real-world workloads, and then roll it out publicly as a commercial offering. Friday AI’s mission with Adaptive Plugin is to make large-scale GenAI efficient, affordable, and sustainable for any organization that relies on long-context LLM operations. Adaptive Plugin isn’t just an add-on. It’s the layer that will make large-scale GenAI financially and operationally sustainable.

13

The private markets are driven by context that never gets captured. Enter context graphs. The team at Foundation Capital (proud former LP here) recently articulated what they call the context graph opportunity: a 'living record of decision traces stitched across entities and time so precedent becomes searchable.' Context graphs allow us to reimagine the future of private markets—not by asking people to retrospectively explain decisions, but by capturing how sourcing, diligence, and portfolio monitoring actually unfold. The patterns that emerge from repeated activity, especially in our world of non-standardized data and edge cases, create a foundation for learning and better investment judgment that doesn't fade with time. This is what we’re building at Unlimited.ai, Inc.. Not just a system of record for transactions, but an agentic system of record for decisions and context: preserving the nuance that would otherwise be lost to time. Today, investment decision making is mostly manual. The opacity and fragmentation of contextual data makes it difficult to learn from past decisions or even to understand how decisions were originally determined. The multi-faceted reasoning behind critical decisions—why you sized that position differently, what made those terms acceptable, why one team was preferable to another—fades into the spreadsheet or inbox abyss once an investment is made. Precedent lives in our heads. And in an era of increasing talent mobility, institutional knowledge has become institutional risk. Automation and orchestration of the workflows, a dream just a few years ago, is becoming table stakes. But you must also capture and integrate data context across workflows….from access to exit. Patchworks of single point solutions that don’t communicate context effectively will cease to compete in the future. The context graph, surfaced through expert agents like our LAILA, allows invaluable insights to emerge, and is grounded in real activity no longer existing solely in our memory. This is the future of the private markets. Where orchestration forms context, and context compounds into alpha.

36

A single plugin can move markets. Models / APIs → Workflows / Skills → Solutions / FDEs Foundation model companies are no longer just “plumbing.”  When the model owner ships the workflow, value migrates up the stack. Anthropic’s launch of legal plugins for Claude wasn’t just a product update — it was a signal. Public legal and information-services companies sold off sharply, as investors repriced what had been viewed as “AI beneficiaries” into potential displacement targets. ▶️ So what changed? /review-contract /triage-nda /vendor-check  /respond Claude introduced legal-specific plugins that automate real workflows: contract review, NDA triage, compliance checks, legal research assistance. These aren’t demos — they’re opinionated, end-to-end workflows, shipped directly into enterprises and bundled with the core model subscription: ▶️ Why the market reacted? ➖ Proprietary data alone is no longer a sufficient moat when competition shifts to the workflow layer ➖ The “ChatGPT for X” playbook just got much harder if foundation model companies ship vertical products themselves ➖ Investors may be pricing in a future where model vendors move beyond APIs into full solutions — potentially even with forward-deployed teams For founders and VCs, this forces a reset. The key question is no longer “does this use AI?”.  It’s “what actually prevents the model provider from building this natively?” Overreaction — or the start of a structural shift? Do model companies stop at tools, or move decisively into solutions?

68

18 Comments

Fundamental opportunities worth solving for in AI right now: Predictability of reasoning, planning, execution of AI. Efficiency and ease of use of running AI workloads. Multiple system integrations - from DBs and APIs to infrastructure and biz systems. Trust, security, guardrails, authorization and auth. Business/science domain specific data, workflows, and optimization. AI cost understanding, control, and reduction.

30

5 Comments

To my dear VC friends, We are all very excited about the growth rates of AI companies. But when you declare "Triple–Triple–Double–Double–Double (T2D3) is dead," you create chaos in your own firm and with your LPs. And your IR team pays the price. 𝗛𝗲𝗿𝗲'𝘀 𝘁𝗵𝗲 𝗽𝗿𝗼𝗯𝗹𝗲𝗺: IR teams just spent 24 months explaining LPs that albeit their portfolio lack liquidity, the fundamentals of their existing portfolio companies are solid, etc... LPs committed hundreds of millions based on that narrative. Now the partners, from the same firm, are saying that model is outdated.... 𝗧𝗵𝗲 𝗿𝗲𝘀𝘂𝗹𝘁? Confused LPs asking: - "So what are the paper returns marked at 3x MOIC on your Saas companies really worth ?" - "Why did you say that the existing portfolio companies were doing well?" - "Why should we trust the new narrative?" - "What else will change next quarter?" IR teams scrambling to explain: --> Why yesterday's pitch is today's problem --> How the new model is totally different (but also kind of the same) --> Why this pivot strengthens the strategy (somehow) 𝗧𝗵𝗲 𝗿𝗲𝗮𝗹𝗶𝘁𝘆 𝗻𝗼 𝗼𝗻𝗲 𝘄𝗮𝗻𝘁𝘀 𝘁𝗼 𝗮𝗱𝗺𝗶𝘁: T2D3 isn't just a growth model. It's a framework LPs have internalized for 10 years on how to evaluate B2B Software companies. They use it to model returns. They use it to compare companies and portfolios. They use it to explain venture investing to their boards and ICs. When you suddenly declare it dead, you're not just changing metrics. You're breaking the narrative that built trust. 𝗧𝗵𝗲 𝗳𝗮𝗹𝗹𝗼𝘂𝘁: → Every fundraise gets harder → LPs hesitate on re-ups and do more due diligence on the portfolios → New LPs question your consistency → IR teams on calls with LPs 10 hours a day and burn out managing the confusion 𝗧𝗵𝗲 𝘀𝗺𝗮𝗿𝘁𝗲𝗿 𝗮𝗽𝗽𝗿𝗼𝗮𝗰𝗵: Evolution, not revolution. Show how the market is expanding. Explain why multiple models can coexist. Build on what LPs understand, don't demolish it. Because here's what matters: LPs need clarity, not contradictions. They need consistency, not pivots. They need to trust your story won't change with the wind. The future of venture depends on this. You can celebrate new growth models without torching the frameworks that built LP confidence in the first place. Your IR team will thank you. Your LPs will trust you. And your next fundraise might actually be smoother. ❤️

14

2 Comments

Why Retention is Your Most Important Metric Most founders obsess over the wrong retention numbers. They look at Day 7 or Day 30 retention percentages and panic when they see single digits. They assume low retention means product failure and pivot prematurely. But this misses the most critical insight about early-stage product development. Retention isn't just about percentages. It's about behavior patterns among the users who stick around. The OpenAI Example Sam Altman almost killed ChatGPT because early retention was "atrocious." But among the small group who retained, usage increased over time. His insight: "If you have a product that has any retention at all, you're in good shape. The default is almost always straight line to zero." The Retention Fallacy The conventional wisdom goes like this: high retention rates indicate product-market fit, low retention rates indicate you need to pivot. This binary thinking has killed countless promising products. The reality is more nuanced. A product with 5% retention where retained users dramatically increase their usage over time often signals stronger product-market fit than a product with 30% retention where usage remains flat. Why? Because the first scenario indicates you've created genuine, growing value for a specific user segment. The second suggests broad but shallow appeal - which rarely survives competitive pressure. What Retention Actually Tells You Retention data contains multiple signals: Immediate retention tells you about initial product appeal and onboarding effectiveness. Low numbers here might indicate friction, not fundamental product problems. Retention behavior patterns tell you about value creation. Are retained users becoming more engaged or gradually churning? This is the signal that matters most. Retention depth tells you about competitive moats. Users who dramatically increase engagement over time are building habits and workflows that competitors can't easily disrupt. The Two Types of Retention Passive retention: Users return but engagement stays constant or declines. They're tolerating your product but not finding increasing value. Active retention: Users return and increase their engagement over time. They're discovering new value and building deeper habits. Most retention analysis focuses on the first metric when the second is far more predictive of long-term success. How to Analyze Retention Correctly Instead of fixating on overall percentages, track: Cohort progression: Among users who return after their first session, how does usage evolve over weeks and months? Feature depth: Are retained users engaging with more sophisticated features or use cases over time? Session intensity: Are retained users spending more time per session or returning more frequently? Workflow integration: Are users building your product into increasingly complex workflows? #Startups #ProductManagement #Retention #PMF #Founders #Metrics #UserEngagement #SaaS Kalaari Capital

22

4 Comments

Worth paying attention to... Anthropic vs. Cursor. While Cursor was paying Anthropic hundreds of millions in API fees to power its AI code editor, Anthropic seems to have studied exactly what developers wanted, then launched Claude Code to "eat" their biggest distribution partner. The result. Claude Code hit $1B ARR in just six months and has already surged past $2.5B, proving that when you're a "wrapper" startup, you're often just paying your provider to learn how to replace you. Keep an eye on Cursor layoffs...

89

13 Comments

Key takeaways: - The core shift is economic, not hype. LLMs turn “AI” into a recurring operating cost, so pilots quickly become capacity planning, governance, and ownership. - Power and timelines show up because production inference needs sustained compute capacity. When usage increases across the org, you run into procurement, security review, integration work, and infrastructure readiness challenges. - The winners are not the teams with the flashiest demos. The winners are the teams that convert compute into measurable outcomes inside a workflow with traceability and controls. - In 2026, “AI strategy” that is not tied to a specific workflow, KPI, and operating model is mostly theater. If you are building the application layer on top of the datacenter buildout, execution systems, workflow automation, data-layer tooling, compliance automation, I want to see what you are shipping. Founder intro: www.axion.ventures

6

The hottest role in AI startups right now isn’t Forward Deployed Engineers. It isn't GTM Engineers. It’s Deployment Strategists. Decagon calls it an “Agent Product Manager.” Harvey calls it a “Solutions Architect.” Palantir Technologies has had versions of this role for years. And the salaries are climbing fast: - Decagon: $200k–$285k - Palantir Technologies: $120k–$200k - Figma: $150k–$260k - Ramp: $100k–$180k - Harvey: $190k–$260k So who are these people? They are usually pseudo-technical -- CS or engineering majors, or folks with technical work experience. Many come from 2 years in consulting, IB, or PE, then jump into startups to get their hands dirty. They are young, hungry, polished, and comfortable being in front of customers. What do they actually do? They make sure enterprise AI deployments succeed. A $100k+ deal does not survive on a nice pitch or a self-serve onboarding flow. It survives if the customer sees value in the pilot. That means: - Embedding directly with the customer - Designing prompt logic for specific workflows - Working with engineering to align integrations and data flow - Helping exec teams define their AI roadmap - Running feedback loops into product and GTM Why does this role matter so much? Because enterprise AI is messy. Integrations, data transfer, and adoption make or break a deal. Most buyers are using AI for the first time, and each has unique workflows. Deployment Strategists bridge that gap. They own the outcome. They are accountable for making pilots successful, which often means millions in revenue down the line. At Conversion, Sam Bochner has been leading this work for us. We are now thinking about scaling it into a full team. Because a few successful pilots can fund an entire department, and the cost of failed deployments is too high to ignore. Is this just a rebrand of customer success? Not really. Success is about answering tickets and renewals. Deployment Strategy is about going deep with a few enterprise accounts, extracting maximum value, and ensuring the pilot closes into a multi-year contract. Call it Agent PM, Solutions Architect, or Deployment Strategist. Whatever the title, this is becoming one of the most important roles in AI SaaS.

679

93 Comments