In partnership with
Google admits Gemini trails on coding, ships 3.6 Flash as a stopgap 🤖⚡
TL;DR
On Alphabet's Q2 2026 earnings call, CEO Sundar Pichai acknowledged Google is behind on agentic coding and confirmed that Gemini 3.5 Pro remains delayed due to coding performance issues. Google shipped Gemini 3.6 Flash one day before the call, scoring 49% on DeepSWE (up from 37% with 3.5 Flash) while cutting output tokens by 17%.
Key Takeaways
Pichai said Google lacks a developer-facing product that generates the usage data competitors already collect, meaning no feedback loop for agentic coding improvements.
Gemini 3.5 Pro, announced at I/O in May, missed its planned June ship date after a training data update failed to fix coding problems. No confirmed release date exists.
Two senior researchers departed in June: Noam Shazeer (co-lead of Gemini) left for OpenAI, and John Jumper (AlphaFold) left for Anthropic, both citing concerns about Google's AI coding position.
Gemini 3.6 Flash produces 17% fewer output tokens than 3.5 Flash, reducing costs while improving coding benchmark scores from 37% to 49% on DeepSWE.
Pichai stated a larger Gemini 4 base model is essential to staying competitive going forward.
Why It Matters
Google publicly admitting a competitive gap in agentic coding is significant. The company that pioneered the Transformer architecture is now playing catch-up in one of the highest-value AI application categories. The talent departures to OpenAI and Anthropic compound the problem, removing institutional knowledge at a moment when Google needs to accelerate.
For builders and operators, the 3.6 Flash release signals that Google is prioritizing cost efficiency and incremental coding gains while the flagship model remains stuck. If you're choosing models for production workloads, the competitive dynamics between Google, OpenAI, and Anthropic on coding tasks are shifting fast, and the gap Pichai described means pricing pressure could intensify as Google tries to win developers back.
Your voice. Every platform. No writing required.
You ghost your own socials by Wednesday. SureThing learns your voice and ships native posts to LinkedIn, X, Instagram, and TikTok, without you writing a thing.
📰 In the News
Headlines & Launches 📣
Cursor, the AI coding company being acquired by SpaceX in a $60 billion all-stock deal, launched a CFO Council with nine confirmed finance leaders from companies including Natera, Payoneer, SentinelOne, and Asana. The council will meet quarterly to build shared productivity benchmarks and a framework for measuring "return on intelligence." COO Jordan Topoleski said the initiative started with a single LinkedIn post and drew an unexpected volume of responses. First session is scheduled for Aug. 18 in San Francisco.
ChatFeatured closed an oversubscribed $2 million pre-seed round, expanding from an initial $1.5M target after investor demand exceeded expectations. Led by Storytime Capital with participation from Garage Capital and BY Venture Partners, the platform helps marketing teams convert visibility data from AI-driven search (ChatGPT, Perplexity, Gemini) into published content and trackable results. If AI search continues eating into traditional search traffic, this is a bet that marketers need a dedicated measurement and response tool.
follow Inc announced followOS, an autonomous AI-powered ad platform that handles campaign strategy, creative production, media buying, optimization, and reporting without an agency or in-house team. A business owner sets an objective, budget, and audience, and the system executes across Meta, TikTok, YouTube, and Instagram. General availability is targeted for Q1 2027 with early access registration open now. Pricing and technical architecture details have not been disclosed.
Hot New Tools 🧰
The vibe coding market hit $4.7 billion in 2026, growing roughly 38% annually, with projections of $12.3 billion by 2027. Lovable reports that 63% of its users have never written code, and founders are its largest user group. Stack Overflow's 2025 survey found 37% of U.S. developers are vibe coding specifically. The catch: the WEF's Future of Jobs Report puts analytical thinking, not coding tools, at the top of employer demand through 2030.
A financial services company spending $14,000 to $18,000 annually on software subscriptions built their own solution using AI-assisted development. The first build took two months but eliminated the subscription cost entirely, running on a server under $80/month for 60-70 users. Version two was rebuilt from scratch in under three days with improved security, authentication, and architecture. The implied threat to the SaaS market is real if thousands of similar businesses follow this pattern.
Research & Innovation 🧪
Gartner predicts 40% of agentic AI projects will be canceled by 2027, not because models are broken, but because governance infrastructure was never built for autonomous agents at scale. This playbook covers the pilot-to-production gap for platform teams in financial services and government, where senior engineers are already spending hundreds or thousands of dollars in tokens daily with positive ROI. The challenge: scaling from successful pilots to thousands of developers without creating compliance exposure.
Miscellaneous 🎁
GitHub Copilot and similar AI coding tools have moved to usage-based billing, converting what was a fixed seat cost into a variable expense with no natural owner in the org chart. Finance built forecasts around numbers that don't move. Engineering controls the usage. Neither team was set up for the gap between them. If your team uses any AI coding tool on a consumption model, someone needs to own the monitoring and set usage thresholds before finance notices overages at quarter end.
Thanks for reading,
— Cagri Sarigoz
P.S. Don't Miss Out on These Resources:
🤝 Follow me on X and LinkedIn for regular updates and insights.
