DevelopingUpdated 3d ago
Fortune 500 pilots are stalling not on model quality but on questions security teams can't yet answer — what an agent did, why, and with whose permissions.
Ryan Marshall · · 6 min read
· 9 min read
· 14 min read
General-purpose model obligations are now live. Most of the panic is misplaced — but three requirements deserve real attention from US teams.
· 8 min read
· 5 min read
· 7 min read
General-purpose model obligations are now live. Most of the panic is misplaced — but three requirements deserve real attention from US teams.
Maya Chen · 8 min read
The cheapest capable model today costs less per token than logging did in 2023. Who absorbs the margin collapse, and where the money went instead.
Ryan Marshall · 5 min read
New deployment surveys show a split stack: frontier APIs for hard problems, self-hosted open weights for everything else.
Maya Chen · 7 min read
A single appellate question — whether ingestion is itself infringement — will shape licensing markets more than any settlement so far.
Ryan Marshall · 9 min read
Two years of mitigations have made attacks harder, not impossible. The honest framing is risk management, and vendors selling certainty deserve scrutiny.
Ryan Marshall · 12 min read
Datacenter capacity grew faster than inference demand this spring. The correction will be uneven — and revealing.
Maya Chen · 9 min read
When every release tops a leaderboard, the leaderboard stops being the story. How to read a model announcement in 2026.
Maya Chen · 10 min read
We tested quantized small models across five phones. Capability is real, battery cost is real, and the marketing gap is measurable.
Maya Chen · 14 min read
Two years of mitigations have made attacks harder, not impossible. The honest framing is risk management, and vendors selling certainty deserve scrutiny.
Why it matters: Companies are deploying agents faster than they are building controls around them.
Ryan Marshall ·
Datacenter capacity grew faster than inference demand this spring. The correction will be uneven — and revealing.
· 9 min read
· 10 min read
Loops, tools, and guardrails: the mechanics behind the year's most-used word, without the vendor gloss.
Why it matters: You cannot evaluate agent products — or their risks — without knowing what is actually inside the loop.
Ryan Marshall ·
The plain-language guide to context limits, retrieval, and memory, and what million-token claims do and don't mean for real work.
· 8 min read
· 7 min read
We tested quantized small models across five phones. Capability is real, battery cost is real, and the marketing gap is measurable.
Maya Chen · 14 min read
Four popular research tools, one messy real-world source pile, and a scoring rubric built around the failure that matters most: confident misattribution.
Ryan Marshall · 13 min read
Twice weekly · Free
The most important AI developments, clearly explained.
No spam. Unsubscribe anytime.
