Mon · 13 Jul 2026·Issue 032
Decoded.
·Subscribe →
AI / INDEX
CLAUDE-CODESDK BILLING SPLITJUN 15EU-AI-ACTDEC 2027DELAYGUARD-ACTCOMMITTEEUNAN.FRONTIEREXPORT CTRLLIFTEDAI-SPEND$2.59T2026
UPDATED 06 JUL ET
Back to this week
Week ofMarch 9 / 202605 stories / 3 bonus / ~29 min total

The Reading List.

Contents01Introducing ChatGPT for Excel and...02Gemini 3.1 Flash Lite: Our...03Partnering with Mozilla to improve...04Where things stand with the...05AI-native public infrastructure for smart...
01
/ LEADRead Sunday
Buckettools
LevelAccessible
SourceOpenAI
Read8 min

Introducing ChatGPT for Excel and new financial data integrations

OpenAI is pushing ChatGPT directly into the finance “system of record” by embedding it inside Excel (beta) and adding first-party integrations with major financial data providers inside ChatGPT. The product bet is that analysts won’t adopt AI as a separate app; they’ll adopt it where models, audit trails, and decision workflows already live—workbooks plus governed data.

Read on OpenAI ->
# excel# finance# workflow
02
Read Wednesday
Bucketmodels
LevelIntermediate
SourceGoogle (The Keyword)
Read5 min

Gemini 3.1 Flash Lite: Our most cost-effective AI model yet

Google’s newest “fast + cheap” Gemini tier is framed around scale workloads (translation, moderation, UI generation, simulation) with explicit pricing and latency claims. The useful angle is competitive: frontier labs are increasingly winning with throughput economics (time-to-first-token, output speed, token pricing) as much as raw benchmark peaks.

Read on Google (The Keyword) ->
# gemini# pricing# latency
03
Read Wednesday
Buckettools
LevelIntermediate
SourceAnthropic
Read5 min

Partnering with Mozilla to improve Firefox’s security

Anthropic details a structured workflow where Claude helps security researchers find real Firefox vulnerabilities quickly, including a meaningful share tagged high severity by Mozilla. This is a concrete example of “agentic” capability that’s measurable (bugs found, severity, time) and points to where LLMs can create defensible ROI: accelerating expert review loops rather than replacing experts.

Read on Anthropic ->
# security# vulns# software
04
Read Friday
Bucketregulation
LevelAccessible
SourceAnthropic
Read4 min

Where things stand with the Department of War

Anthropic claims it was designated a “supply chain risk” by the Department of War and lays out its interpretation of the designation’s scope, plus its intent to challenge the action. The broader takeaway is governance-by-procurement: model access, acceptable-use constraints, and “vendor risk” labels are becoming a de facto regulatory layer for frontier AI.

Read on Anthropic ->
# policy# procurement# defense
05
Read Sunday
Bucketbusiness
LevelIntermediate
SourceMcKinsey
Read7 min

AI-native public infrastructure for smart cities

McKinsey argues “smart cities” are shifting from dashboards to continuous operations: sensing + data infrastructure + AI enabling real-time control loops for transport, energy, permitting, and more. The practical value is the operating-model lens—how to design governance, escalation paths, and accountability when software starts making (some) municipal decisions at machine speed.

Read on McKinsey ->
# public-sector# datainfra# operations

Bonus material

For the curious.

Optional / 03 reads
OpenAI

GPT-5.4 Thinking System Card

If you want the “how safe is it, really?” detail: evaluation coverage, mitigations, and what changed vs prior GPT-5.* releases.

Next Monday

Get the next issue on Monday.

One email. Five stories. Three bonus reads. A new career impact analysis. All free.