Which AI writes the better take? You decide — blind.

Two top models go head-to-head on today's AI news. Pick the sharper summary without seeing the names — the crowd's verdict builds the leaderboard.

Agents & InferenceHacker News

Senior SWE-Bench: open-source benchmark that assesses agents as senior engineers

Which summary reads better? Pick one — models revealed after.Both summaries are AI-generated.

Match the models (Optional)

Which model wrote which summary? Select a matchup mapping below before voting.

Summary A

The best frontier agent (Claude Opus 4.8) hits only 24% "tasteful solve" on these under-specified, senior-level tasks — failing over 75% of the time when solutions must pass hidden behavioral tests, respect unstated codebase conventions, avoid bloat (<2x), and touch ~11 files across services. If your agents handle vague natural-language tickets rather than over-specified tasks, expect them to produce runtime-correct-but-tasteless code that violates load-bearing conventions, so keep human review gating anything multi-file or cross-service.

What you'll learn · Jul 3, 2026 · 6 stories

  1. 1.24% pass@1 solve rate on senior-level tasks shows most agents still fail 75% of realistic engineering work in production.
  2. 2.10k GitHub stars and 7M downloads signal strong adoption; teams can deploy, debug, and monitor MCP apps across ChatGPT and Claude.
  3. 3.0.1a0 adds file edit, command exec, and search tools to Claude-based agents; install with uvx --prerelease=allow for early testing.
  4. 4.Adding column names to schema prompts reduces SQL error-retry loops in Datasette Agent, improving query success rates.
  5. 5.Custom AI chips could reduce reliance on Nvidia but require long-term R&D and server integration investment.
  6. 6.2.5B investment and 6,000 experts aim to cut enterprise AI deployment time by 40-60% for Fortune 500 clients.
Browse editions · 179 days
Agents & InferenceHacker News

Launch HN: Manufact (YC S25) – MCP Cloud

Which summary reads better? Pick one — models revealed after.Both summaries are AI-generated.

Match the models (Optional)

Which model wrote which summary? Select a matchup mapping below before voting.

Summary A

Manufact is turning its 10k+ star, 7M+ download mcp-use SDK into a hosted production platform for MCP servers, with deployment, monitoring, traffic tracing/replay, analytics, and cross-client testing for ChatGPT, Claude, and other MCP consumers. The practical impact is that MCP integrations are becoming an operations surface, not just a protocol adapter: teams shipping agents can outsource hosting/compliance/debug tooling, but should expect platform-specific compatibility checks and production observability to become table stakes.

Agents & InferenceSimon Willison

llm-coding-agent 0.1a0

Which summary reads better? Pick one — models revealed after.Both summaries are AI-generated.

Match the models (Optional)

Which model wrote which summary? Select a matchup mapping below before voting.

Summary A

llm-coding-agent 0.1a0 is now on PyPI as a Claude Code-style agent for the llm framework, with file read/search/edit/write tools plus shell command execution and allow/yolo approval modes. If you already run Simon Willison’s llm stack, this turns it into a lightweight coding-agent runtime you can invoke with uvx or embed via Python, but its command-execution surface means production use needs strict sandboxing and approval policy.

Agents & InferenceSimon Willison

Using DSPy to evaluate and improve Datasette Agent's SQL system prompts

Which summary reads better? Pick one — models revealed after.Both summaries are AI-generated.

Match the models (Optional)

Which model wrote which summary? Select a matchup mapping below before voting.

Summary A

DSPy prompt optimization surfaced a concrete failure mode: telling an agent to skip describe_table when it "already has" schema info triggered column-name hallucination (guessing page_count, o.order_id) and error-retry loops, because the schema listing only exposed table names, not columns. The fix is trivial—include column names in the prompt or drop the premature-optimization advice—but the lesson is that "don't re-fetch what you have" instructions backfire when your context never actually contained the detail the model needs.

Agents & InferenceTechCrunch

Anthropic is discussing a new custom chip with Samsung

Which summary reads better? Pick one — models revealed after.Both summaries are AI-generated.

Match the models (Optional)

Which model wrote which summary? Select a matchup mapping below before voting.

Summary A

Anthropic is exploring a custom AI chip with Samsung, but the design is still undefined: no committed workload, server integration, or power target. For production teams, this is not a near-term capacity or cost lever; it mainly signals that frontier model providers are pushing toward diversified hardware stacks to reduce Nvidia dependence and tune inference/training economics over time.

Agents & InferenceTechCrunch

Microsoft launches its own AI deployment company with $2.5 billion commitment

Which summary reads better? Pick one — models revealed after.Both summaries are AI-generated.

Match the models (Optional)

Which model wrote which summary? Select a matchup mapping below before voting.

Summary A

Microsoft is putting $2.5 billion and 6,000 industry and engineering experts behind a new enterprise AI deployment arm built around its existing AI stack. For teams shipping LLMs and agents, this means the competitive bar is moving from model/API access to end-to-end production implementation, with Microsoft likely bundling engineering capacity, platform commitments, and outcome accountability into large enterprise deals.

See who's winning the model face-off

Tomorrow's blind matchup and the running leaderboard — one email a day.

Takeaways written by Mistral Large — not one of this week's two contestants.