- Everyday AI
- Posts
- Ep 677: The 3 Big Obstacles Holding AI Adoption Back
Ep 677: The 3 Big Obstacles Holding AI Adoption Back
Meta working on Nano Banana competitor, OpeAI rolls out GPT-5.2 Codex, DOE launches Project Genesis and more
Everyday AI
December 19, 2025
š Subscribe Here | š£ Hire Us To Speak | š¤ Partner with Us | š¤ Grow with GenAI
Outsmart The Future
Today in Everyday AI
8 minute read
š Daily Podcast Episode: Learn the 3 major obstacles holding enterprises backāfrom infrastructure to data gapsāand how to overcome them for true business transformation. Give it a watch/read/listen.
šµļøāāļø Fresh Finds: Gemini officially gets NotebookLM integration, Claude rolls out its Chrome extension to more, Adobe and Runway partner and more. Read on for Fresh Finds.
š Byte Sized Daily AI News: Meta working on Nano Banana competitor, OpeAI rolls out GPT-5.2 Codex, DOE launches Project Genesis and more. Read on for Byte Sized News.
šŖ Leverage AI: Jeetu Patel, the Presidnet of Cisco, lays out the AI implementation playbook. Keep reading for that!
ā©ļø Donāt miss out: Did you miss our last newsletter? ChatGPTās App Store launches, Lovable raises $330 million, Amazon makes big AI switch and more. Check it here!
EP 677: The 3 Big Obstacles Holding AI Adoption Back -- An Everyday AI Chat with Cisco President Jeetu Patel
Jeetu Patel knows a few AI secrets.
As the President of one of the largest companies in the world, he's helped pave the AI adoption roadmap.
At Cisco, they provide full-stack, enterprise AI solutions spanning infrastructure, security, observability, and operations to the world's largest companies.
So naturally, Jeetu could write a legit playbook on what's slowing enterprises down in the AI fast lane and how they can overcome those bottlenecks.
And naturally, Jeetu is gonna share it all with us.
Also on the pod today:
ā¢ The $5T data center boom šø
ā¢ Why OpenAI hikes prices š
ā¢ Are we running out of data? šāāļø
Itāll be worth your 33 minutes:
Listen on our site:
Click to listen
Subscribe and listen on your favorite podcast platform
Listen on:
Hereās our favorite AI finds from across the web:
New AI Tool Spotlight ā Loki is an AI-powered landing page builder, Userology is an AI-backed product researcher, Vurge brings together AI, web scraping and Google Sheets.
Gemini and NotebookLM ā After a series of rumors and leaks, Google officially confirmed and launched NotebookLM integration in Gemini.
ChatGPT Interface ā ChatGPT has rolled out pin chats, allowing users to better organize their chats.
AI Videos ā Adobe and Runway partner for more AI video utility.
Claude Chrome ā Claude is opening up its Chrome extension for all paid plans, including Plus users.
AI Models ā OpenAI finds chain-of-thought monitoring usually beats output-only checks, and probing longer or with follow-ups reveals hidden intentions. Curious how model size and a āmonitorability taxā change the game?
Anthropic Agents ā Anthropic open-sourced its Agent Skills standard and a partner skills directory, betting that shared, portable "skills" will become the backbone for enterprise AI assistants.
AI Pilots ā AWSās Generative AI Innovation Center helped 1,000+ companies move generative AI from pilot to production, with over 65% of projects going live. Want to see how?
1. OpenAI rolls out GPTā5.2āCodex with bigger coding and cyber chops š ļø
According to OpenAI, GPTā5.2āCodex is now available to paid ChatGPT users and brings stronger longāhorizon coding, more reliable tool use, and better handling of large refactors and migrations, making it more dependable for extended engineering work.
The model posts modest benchmark gains but shows notable improvement in terminal and cybersecurity tasks, where OpenAI says captureātheāflag testing reveals a clear capability jump over prior Codex releases. OpenAI is moving cautiously on access, piloting a trusted partner system and inviteāonly programs for defensive security researchers while planning API availability soon.
2. Report: Meta to compete with Googleās Nano Banana with new image model š„
Meta says it will roll out a new text-focused large language model called Avocado and an image-and-video model code-named Mango, with releases expected in the first half of 2026. Alexandr Wang, Metaās chief AI officer, told employees the Avocado effort will prioritize better coding ability and that the company is exploring visual "world models" to teach AI about environments.
The moves follow a summer restructuring that created Meta Superintelligence Labs and a hiring sweep of researchers from OpenAI, underscoring Metaās push to compete on multimodal AI. Metaās recent Vibes video generator and the fast follow-up from rivals show the battle for leadership in image and video AI is already active and accelerating.
3. Cursorās Anysphere buys Graphite to stitch writing and review into one AI coding pipeline š¤
Anysphereās Cursor is acquiring code-review startup Graphite in a cash-and-equity deal that keeps Graphite independent but promises deeper integration, a move announced as the transaction nears closing in the coming weeks.
The acquisition targets a growing bottleneck: AI tools have sharply sped up code writing, while human review processes have not kept pace, and Cursor says combining its editor with Graphiteās review tooling will speed safe, context-aware shipping. Both CEOs frame the deal as building an end-to-end AI coding platform that preserves review quality for AI-assisted code, while keeping both products separate through 2026 as integrations roll out.
4. Microsoft and Google Join DOE, Plus 22 More, to Supercharge AI Energy Research ā”
The Department of Energy today announced partnerships with Microsoft, Google and 22 other major tech firms to officially launch the Genesis Mission, a national effort to use AI to automate experiments, speed simulations, and build predictive models for renewable energy, manufacturing, and medicine.
The initiative pulls cloud and chip leaders together with national labs to boost U.S. applied AI competitiveness and create a shared research platform. DOE and White House science officials called the agreements the start of broader public-private collaboration to accelerate scientific discovery. In short, Genesis shifts the federal role toward orchestrating industry-wide AI resources for faster, more efficient breakthroughs.
5. U.S. begins review to possibly allow Nvidia H200 chip sales to China šØš³
The Commerce Department has sent license applications for shipments of Nvidiaās H200 AI chips to the State, Energy and Defense Departments for a 30-day interagency review, marking a timely move after President Trump signaled he would permit those sales with a 25% fee, sources say.
The decision would reverse stricter Biden-era export limits on advanced AI chips and could ease Chinaās access to powerful AI hardware while the administration argues it keeps U.S. firms technologically dominant by undercutting Chinese chipmakers. Critics warn the chips could strengthen Beijingās military AI capabilities and erode U.S. advantages, while supporters in the administration say wider sales might discourage Chinese efforts to catch up.
š¦¾How You Can Leverage:
Why's the AI math not mathing?
More than 90% of enterprise leaders say AI adoption is a top priority but less than 10% have actually deployed it company-wide.
Huh?
That gap ain't an accident, y'all. There's three specific bottlenecks killing implementation while your competitors figure out the workaround.
Jeetu Patel, President and Chief Product Officer of Cisco, showed up to Everyday AI with the dang playbook.
As the executive running AI infrastructure, security, and operations for some of the world's largest enterprises, Jeetu's seen every failure pattern and knows exactly what separates the 10% winning from the 90% stuck.
His insights?
On point, like a decimal.
Jeetu broke down the three big bottlenecks blocking AI adoption, revealed why data centers are being built where power exists instead of where companies need them, and explained how smart executives are securing competitive advantages right now.
Wanna know the future of enterprise AI? Learn from someone literally building it.
Let's get it.
1 ā Infrastructure Wars: Secure Capacity or Watch From Sidelines š„
$20 monthly? Lost money. $200 monthly? Still bleeding cash. Next up? $2,000 plans, then $20,000.
That's not broken pricing, fam. That's infrastructure scarcity showing you what real enterprise demand looks like when supply can't keep up.
Jeetu's take on this is wild: Data centers are being built where power EXISTS, not where companies actually need them. Token generation is becoming national security infrastructure because countries figured out the economic prosperity connection.
The $5 trillion question? Who secures capacity first.
Here's Jeetu's warning: Companies that started experimenting early are finding what works. The ones waiting for "perfect technology" are falling further behind every quarter.
His move? Start now with whatever infrastructure you can get, because everyone's competing for the same limited GPUs and power grids
Try This
Call your data center provider Monday and ask about power availability and GPU access for Q3 2025āactual timelines, not sales talk.
Map three workflows that could run autonomously if compute wasn't the constraint. That's your real roadmap.
Project your AI spend 18 months forward at 3x growth. Show your CFO before budget season locks you out.
Pick the business unit with highest ROI per compute hour and give them capacity first. Stop spreading infrastructure thin across departments hoping something sticks.
2 ā Trust Deficit: Your Models Are More Vulnerable Than You Think š
Large language models are nondeterministic. Same question, different answers every time.
Yet you're building mission-critical business systems on this.
Jeetu's team at Cisco jailbroke DeepSeek 100% of the time within 48 hours using Harm Bench benchmarks. Every vulnerability exposed before companies deployed it.
His method? Trick models to find where safety fails.
Ask a model to build a bomb and it refuses. Ask it to write a movie script where Brad Pitt builds a bomb? Suddenly you've got detailed instructions.
That's a jailbreak, and hackers are way better at this than your security team.
Jeetu's play for developers: Stop building security stacks from scratch. Use guardrail APIs and innovate without paranoia breaking your timeline.
The companies winning on trust? They're validating continuously every time they retrain models, not running annual audits and hoping for the best.
Try This
Spend 30 minutes Thursday trying to break your customer-facing AI. Ask it to generate content your brand would never publish. Document what slips through.
Assign someone to roleplay a malicious user for two hours manipulating outputs through creative prompts.
Six months into AI deployment without algorithmic security testing? You're already vulnerable. Schedule that pressure test this week.
Compare guardrail API costs against building in-house security. Most teams are shocked at the difference.
3 ā The Machine Data Blindspot: 55% of Growth You're Ignoring ā”
Most companies think their proprietary customer data is their competitive moat.
Jeetu's reality check: 55% of global data growth is machine-generated now. Not human content. Machine logs from autonomous agents working 24/7.
We've exhausted publicly available internet data for training models. Models now run on synthetic data and machine logs showing what agents did, when, and what happened next.
That's completely different from human documents and emails your data team knows how to handle.
Jeetu's insight? Correlate machine data with human context for actual differentiation. But most companies organize information like it's 2020, optimized for document search instead of time-series analysis.
His team built infrastructure specifically for machine data streams. The pattern he's seeing? Autonomous execution jumped from 20-minute tasks to 30-hour coding sessions generating massive data streams.
Most companies ain't equipped to capture that value.
Try This
Audit what percentage of your data growth is automated systems versus humans. Most executives don't know this number.
Pick your highest-volume automated workflow. Connect those machine logs to an LLM in a pilot by month-end.
Ask your data team about time-series machine data retrieval Tuesday. If they look confused, you found your gap.
Prioritize AI integration for workflows already generating machine data instead of manual processes needing complete overhauls.
Reply