
founder @ Tribble, ex-GM @Salesforce & @SAP building products. 馃嚭馃嚫 馃嚚馃嚘
San Francisco, CA
Joined February 2009
- Tweets1.8K
- Following1.9K
- Followers861
- Likes7.3K
Love that weve reached the point where solving an Erd艖s problem is now basically a legendary drop in Diablo. Insanely low drop rate, still gated on skill. You can't force the drop, you can only pick the farm and raise your kill rate. Math and physics become loot grinds?
1
2
77
Imagine a map
with no distortions,
no forgotten roads,
no missing grain of sand.
Every ocean, an ocean.
Every mountain, a mountain.
Every life,
still searching for itself
within its own reflection.
At some point,
the distinction disappears.
It is no longer a map.
It is reality.
52
Credential Blast Radius is the agent-era version of 'too much access'
most enterprise teams cant tell you what an agent can touch. its not a permissions list. its an MCP graph 4 tools deep
your eval covers 'does the agent finish the task'. it doesnt cover 'what else can the agent do that nobody asked for'
every prod incident ive seen with agents in 2026 traces back to this one gap
2
1
63
local VLMs hit a wall around structured output
qwen2.5-vl 7b runs at 30+ tok/s on m3 max with mlx. great for descriptions
ask it to return a strict JSON schema with image refs and bounding boxes and the failure rate triples vs the same model on llama.cpp with outlines
mlx 'json mode' is essentially 'try not to break too obviously'. for prod you need outlines or guidance bolted on top
most enterprise teams default to a hosted vendor here because of this exact gap. but the gap closes in 2 PRs. its a tooling problem, not a model problem
2
1
93
Vendor Roulette is the new Silent Churn
your vendor quietly swaps the underlying model. your answers change. your eval was one-shot at pilot and nobody runs it anymore
you dont get notified. you dont get a SLA. you just notice your agent 'feels worse' and spend 3 sprints debugging your own code
the vendor calls it 'continuous improvement'. the buyer is just spinning the wheel
27
DeepSeek V4 just reset every enterprise inference budget
V4 Flash at 284B benches within ~2pt of opus 4.6 on reasoning. V4 Pro at 1.6T goes toe-to-toe with the frontier
thats per-token pricing at 1/10th. self-host on an 8xH100 matches the throughput youre paying $60-80k/mo for via API
every RFP with '2025 per-token rates' is now a renegotiation waiting to happen
the hard part isnt the model. its telling legal you want to self-host a chinese-origin weights file in prod
1
74
the fix isnt better vendor trust. its continuous eval on your own workload. run the same 100 prompts every monday, alert when pass-rate drops 5%. this is 2 hours of work that most enterprise ai teams still wont build
40
'scale to mass intelligence' reads as cover for 'we figured out how to train cheaper than anthropic'. the interesting eng isnt the 862B pro model, its the 158B flash that gets within 2pt of opus 4.6 on reasoning. thats the enterprise deployable
36
cheap and frontier-tier is fine for devs. for enterprise buyers the bigger question is SLAs and data residency on a china-hosted model. most regulated industries cant touch V4 without a self-host plan. which is where unsloth-style tooling gets interesting
30
a 284B Flash that benches near opus 4.6 max is the enterprise inference bombshell. anyone with an 8xH100 or 8xMI300X can self-host frontier-tier without burning $80k/mo on API. the self-host ROI math just flipped again
38
26 tool calls without rails falling off is the real number here. most enterprise agent pilots die at 3-4 tool calls because the eval never covered the 5th. unsloth showed what a tuned eval loop looks like
1
1
30
the enterprise version is smaller and already here. procurement portals, sales intake forms, CRM notes. one poisoned document in the retrieval index and your agent makes decisions the attacker wrote. most RAG deployments are wide open
30
the real story isnt peak tok/s. its that AMD consumer cards are 2x-price competitive with nvidia only when the driver is custom. stock rocm still leaves 40% on the table. most enterprise vendors skip AMD because of that driver gap
59
the time arbitrage is wild. labs compress 6 months of decision into 6 hours. enterprise procurement expands 6 weeks of decision into 6 quarters. vendors who help buyers catch up win the next cycle
26
the Engagement Layer Tax, second time around
sap ran this play when salesforce became the engagement layer over their records in 2015. flipped to api-metered pricing, bills jumped for the same workflow
headless 360 is the setup. the "agent access to records" sku is the punchline. 12 months max
1
2
93
Pilot Purgatory isnt a bug, its the product
6-month pilot. nobody defined "done". the champion rotates. the tool stays in quarterly review forever
the vendor calls it "land and expand". the buyer calls it "still evaluating"
38



